Summary

SUMMARY: Performance "Rework has_trait_flag to check mutation cache instead of all traits, overhaul with support for trait flag strings leading to IDs instead of using std::string"

Purpose of change

Back while I was working on #1486 it was brought up that has_trait_flag evidently taxes performance moreso than has_trait, with this being a blocker.

Since I like JSONizing hardcoded trait effects so that modders can do cool stuff with them, it would be a good idea to make has_trait_flag have comparable performance impact as has_trait. It turned out to be significantly more complicated due to how flags are (or really, aren't) handled under the hood, but the original idea is still buried in there alongside the bigger changes per Coolthulhu's suggestion. In short, mutation flags had to be defined as a whole separate thing that the code can look up in more efficiency ways than the usual method that strings are grabbed for all other flag checks.

Describe the solution

JSON Changes:

• Created flags_mutation.json, being used to set aside a mutation_flag entry for every trait flag currently used by the code. Also went ahead and defined the mycus flag used for dialogue screening of Marloss mutations and the flags used only by cosmetic hair traits.
• Removed NON_TRESH from the regular flags.json, being the only trait flag with an entry there.

C++ Changes:

• In activity_handlers.cpp, set the trait flag definitions used elsewhere in the file to look for trait_flag_str_id instead of mere strings.
• In character.cpp, updated Character::practice so that the checks for the pred trait flags are formatted to work with the new trait flags, doing the same with Character::apply_skill_boost now with removing the self-contained definitions of flags that probably should be moved to the top of the file if they're going to be kept (and might still be, as per alternatives section). Character::mod_thirst, Character::mod_rad, and Character::set_highest_cat_level were all also updated so that uses of has_trait_flag in them all asked for trait_flag_str_id instead of a raw string.
• In character.h, updated its definition of has_trait_flag to ask for trait_flag_id instead of std::string.
• In condition.cpp, set the set_has_trait_flag check so that it feeds trait_flag_to_check to has_trait_flag as a trait flag string instead of just a string.
• In consumption.cpp, set Character::will_eat so that its use of a trait flag check uses a trait flag string instead of a string.
• Added flag_trait.cpp, derived from the contents of flag.cpp to define trait flags as something more than just a simple std::string. Per Coolthulhu's advice in this PR and on the discord, this establishes the JSON to be checked and checking that the code is using trait flags that actually exist in the JSON. This is what defines trait_flag_id as IDs for trait flag checks in the code to use, and trait_flag_str_id as strings for trait flags that can be checked and handled in whatever ways it needs that just being fed std::string wholesale won't allow.
• Added flag_trait.h, likewise based of flag.cpp.
• In init.cpp, added include for flag_trait.h, updated DynamicDataLoader::initialize to tell it that mutation_flag is a type of JSON entry that exists, and added relevant checks for flag_trait stuff to DynamicDataLoader::unload_data, DynamicDataLoader::finalize_loaded_data, and DynamicDataLoader::check_consistency.
• In item.cpp, set the definition of a trait flag used in the file to give it a trait_flag_str_id instead of std::string.
• In iuse_actor.cpp, updated checks for trait flag strings in enzlave_actor::use.
• In magic.cpp, updated checks for trait flag strings in spell::spell_fail.
• In melee.cpp, updated checks for trait flag strings in Character::roll_bash_damage, Character::roll_cut_damage, and Character::roll_stab_damage.
• In mondeath.cpp, updated checks for trait flag strings in mdeath::guilt.
• In mutation.cpp, updated Character::has_trait_flag so it asks for a trait_flag_id instead of std::string. In addition, as per the original focus of this PR, it's set to check cached_mutations instead of calling get_mutations, with the expectation that this will have some improvement in performance should more uses of traitflag checks be added to the code.
• In mutation.h, set it so that mutations look for trait_flag_str_id instead of std::string when they ask for flags.
• In mutation_data.cpp, set mutation_branch::load so that loading mutation flags correctly tries to read trait flag strings instead of using string_reader. For now I've kept it so the JSON in mutations still says flags instead of changing it, to save some editing of mutation JSON and so modders don't have any mods that add mutations with flags instantly broken by this PR.
• In type_id.h, added a class for json_trait_flag.

Documentation Changes:

• Added all flags in use in the hardcode to JSON_FLAGS.md along with an explanation of them being a different sorta JSON.

Describe alternatives you've considered

• Converting all uses of trait flag strings in the code to pre-defined examples at the top of the file, so they don't have to define the trait_flag_str_id multiple times per file. For now I've kept usage in the files consistent with original usage, this could be done in this PR or a followup, whatever would be preferred.
• Setting it so that mutations call for trait flags using something other than flags. This would help reduce confusion about the fact that mutation flags are a whole different type of JSON thing from other flags, but as noted earlier it would increase the amount of minor JSON edits and modders would have to update any mutations they have that use flags. With this, all the work is already done on our end and any flags used by the code will still behave the same for those modders.
• Giving Sapiovore the SAPIOVORE flag and changing whatever trait flag checks we need to to make it actually have an in-repo use in mutation JSON. No idea what we'd even what it to do separately from Cannibal though.
• JSONize all the things!

Testing

• Compiled and load-tested.
• Tested that butchering a dead NPC warns about descrating the dead without relevant mutations, and that this goes away if you have Cannibal, Psychopath, or Sapiovore.
• Tested whaling on a debug monster both with and without Apex Predator, confirmed that focus would tank without it but stay at normal levels with it.
• Also confirmed that punching the debug monster gave 5% to melee skill and 9% to unarmed with the very first hit, while with Apex Predator the amounts for the first hit were 9% and 15%
• Waited around a while with one of the relevant predator traits to confirm the "you recall your most recent hunt" message still shows up.
• Tested temporarily adding NO_THIRST and NO_RADIATION flags to a trait, tested sitting around in a HWS for a while both with and without that trait, confirming that both work as intended.
• Checked that adding Killer Drive incremented Lupine mutation category normally (showed as 8), tested adding the NON_THRESH flag to it and confirmed that it no longer counted as adding to mutation category strength.
• Messed with forcing marloss cultists to spawn, added the Spiritual trait, confirmed that when you talk to them you can pick a fight with them by calling their wacky shenanigans heretical, and that this dialogue option disappears if you have Marloss Carrier, thus confirming that NPC dialogue use of has_trait flag still works.
• Ate some delicious people meat, as intended if you lack the Cannibal trait the warning message notes that eating people is mean.
• Confirmed that item screen for human flesh states the obvious, in red if you lack the relevant trait, in green if you have it.
• Spawned in a zombie and killed it, tested the morale penalty needed to become uncomfortable with neutering it, confirmed that Predator and Apex Predator compensates. Likewise confirmed that having Culler takes a bite out of that 250 morale penalty you'd normally get, while doing this with Apex Predator means no morale hit at all.
• Tested giving myself Rat Claws and wailing on a debug monster. Unarmed hits leaned towards the 10-12 range at zero unarmed skill, 20-22 at 10 unarmed skill. Tested that damage goes back down if the UNARMED_BONUS flag is removed from Rat Claws.
• Tested turning a zombie child into a delicious red smear with mjolnir, PC gets the smol sad if no traits, gets the message about culling the weak with Culler trait (and morale hit is smaller), no message with Predator.
• Recruited starter NPC, confirmed that I could ask them about their backstory without any issues.
• Tested editing the evac merchant so you can't trade with him without a mutation threshold, confirmed that use of trait flag checking also works without issue, and that it's a special case where the trait flag doesn't have to be defined in JSON

Dumb thing: Culler and Hunter reduce the morale penalty from enzlaving a zombie, and Apex Predator eliminates it outright, but PRED3 is in a derpy weird spot where it increases your tolerance for being willing to start mutilating a zombie, but does nothing for how massive a penalty you take from it like its preceding traits do.

I don't actually know how to confirm how much better this performs than using get_mutation...

Additional context