Fix wheels release

[megaserg]

Use latest cibuildwheel and fix all errors.

[megaserg]

@doublethefish this builds all the wheels successfully on my fork :) would appreciate if you could take a look!

[doublethefish]

Oh awesome!

And it only took 32mins to complete on your fork; did you do anything specific to make it take less than the 5 hours I was seeing?

[doublethefish]

It is always better to put more information that "Appease clang" for changes like this.

In future I would like to see the specific "appeasement" that is occurring but in this case I assume it is a warning about template ambiguity.

moved to commit.

[megaserg]

Thanks @doublethefish !

Sorry, I wasn't very diligent with the commit messages and history, I was expecting that you'd squash the whole PR 😌

Regarding clang change: it is to fix this error:

        /Users/runner/work/tmap/tmap/src/_tmap/cereal/types/tuple.hpp:119:85: error: a template argument list is expected after a name prefixed by the template keyword [-Wmissing-template-arg-list-after-template-kw]
          119 |     tuple_detail::serialize<std::tuple_size<std::tuple<Types...>>::value>::template apply( ar, tuple );
              |       

which is a relatively recent addition to clang: llvm/llvm-project#80801

Regarding 32 mins: that number is end-to-end but depends on available parallelism (in GitHub free tier?) at the moment. Each job takes about 5 mins (10 mins for macos x86_64) so best case you'd see 10 mins end-to-end, worst case ~2 hours (if everything is sequential).

Thanks again! I believe some of the long-standing issues can be closed now :)

[doublethefish]

Sorry, I wasn't very diligent with the commit messages and history, I was expecting that you'd squash the whole PR 😌

No worries, thanks for the PR. We are very grateful.

Regarding clang change: it is to fix this error:

Thanks for that it's useful to the specific warning, and edifying to know that I guessed right about the type of warning it was 🤓. As you probably know, -Wmissing-template-arg-list-after-template-kw is a warning that only errors because we have warning-as-errors on for that bit of code. Similar template-specific warnings have been around for a while, and problematic; for instance, at one point fixing this exact issue in one compiler, but leaving spaces between the angle-brackets, on another compiler would cause less-than symbol errors 🤦🏻‍♂️.

[megaserg]

Sounds great @doublethefish ! Would you be willing to release the wheels to PyPI?

[doublethefish]

Yes, of course. But it will have to be sometime over the next few days.