New API for supporting mut aliasing safety

[Indra-db]

This PR introduces:

• locking for either component records (sparse components) or table-column locks (non-sparse components).
• separate per-thread / per-stage locking for table columns.
• everything gated by the FLECS_MUT_ALIAS_LOCKS define, so there is no performance impact when it’s not enabled.
• build tests and unit tests for the behaviour.

End users of this internal API must use it themselves and panic when the mut-alias guarantee is violated.

ecs_lock_target_t holds information about where the lock should be applied. For sparse components/IDs (sparse and/or non-fragmenting) it will use the component_record for locking. For table-column components it will lock the individual column.

ecs_get_id and ecs_get_mut_id have been split, it now calls ecs_record_get_id / ecs_record_get_mut_id which starts from an entity record. The main purpose is that the ecs_record_get_id returns an ecs_get_ptr struct instead of just a void*, which contains the lock target information. The second purpose of this change is that stop the Rust bindings for flecs from maintaining their own get functions (a maintenance hell). Flecs-rust retrieves the record once, then uses its get callback to retrieve component pointers. I believe the same approach could be used for the C++ get callback.

Changes to ecs_get_id and ecs_get_mut_id need to be benchmarked by Sander.

Changing ecs_get_id and ecs_get_mut_id means a lot more scrutiny has to be applied to this PR. This could cause performance regressions that depend on compiler-specific inlining behavior.
I'll need to test this myself — not sure if I'll have time before the next release, but I'll try.

[Indra-db]

ci test, I expect a failure somewhere, I cannot run tests locally due to bake being problematic (see discord)

[Indra-db]

all tests passing 🥳

[Reddy-dev]

ci test, I expect a failure somewhere, I cannot run tests locally due to bake being problematic (see discord)

@Indra-db could you link the message, I think i also have similar issues

[Indra-db]

@Reddy-dev

Tldr, Sander told me to use bake run as opposed to bake test which should fix the ouput

[Indra-db]

@Reddy-dev this was my thread where I was dumping my experience (which I closed)
https://discord.com/channels/633826290415435777/1406452408149741770/1406998533449580585

[Indra-db]

update to latest main

[SanderMertens]

I'm still seeing performance regressions when comparing with latest master:

I think it's better to separate out the implementation into its own functions so there's no risk of regressions.

(Note: benchmarks were done without the FLECS_MUT_ALIAS_LOCKS flag on a regular Flecs master)