Formalize the meaning of fully-qualified type names

[jleibs]

We currently follow a convention of adding name prefixes for:

• rerun.archetypes
• rerun.components
• rerun.datatypes

As of #3660, we no longer add any prefixes to user components.

The rerun.components prefixes, specifically, make it all the way through the data model and would otherwise show up as the component/field names in the UI. As such we have special handling to strip off the rerun.components and to annotate components ending with Indicator differently.

If these conventions are going to carry special meaning we should capture that with some form of RFC.

• What makes a valid name? Are there reserved/illegal characters?
• Do we always want to require an "org" prefix such as rerun?
• These names seem likely to play a role in plugin-handling / dispatch in the future.
• Will there be special handling of the intermediate type organization? Is there a good reason for someone to match, e.g. ".components.".

[teh-cmc]

I really don't like the idea of having semantics attached to fully-qualified names.

It obviously is the case today, but to me that's always been more of a crutch than anything, so we can keep the ball rolling until we can get an actual runtime schema registry with proper schema metadata / extensions / etc running:

• Implement minimal datatype registry for cross-batch deduplication #1809

Once we have this registry (and we're close: #3741), introspecting the schema associated with a fully-qualified name becomes O(1) and queries like:

registry.get_schema("rerun.components.Point3D")?.is_builtin()
registry.get_schema("rerun.components.Point3D")?.is_component()
registry.get_schema("rerun.components.Point3D")?.is_indicator()

become trivial.

Similarly, I would expect we won't be looking for indicator components by parsing their suffixes in the future: there should be a RERUN:indicator marker in their schema metadata or something.

I do think it's perfectly fine to make use of known prefixes for trivial things though (e.g. UI hints), even in a world with a runtime registry, but that's as far as they should go semantic-wise.

Plugin handling and dispatching, as you say, is where I expect fqnames to really play an important role.
Especially once we start opening plugins to everyone, at which point namespacing conventions will become super important.

[teh-cmc]

• Related: https://github.com/rerun-io/rerun/pull/3790/files#r1354214405

• Related: Explicitly annotate all Spatial types as 2D or 3D #3785