Fix MeshAllocator panic

[eero-lehtinen]

Objective

Fixes #14540

Solution

• Clean slab layouts from stale SlabIds when freeing meshes
• Technically performance requirements of freeing now increase based on the number of existing meshes, but maybe it doesn't matter too much in practice
• This was the case before this PR too, but it's technically possible to free and allocate 2^32 times and overflow with SlabIds and cause incorrect behavior. It looks like new meshes would then override old ones.

Testing

• Tested in loading_screen example and tapping keyboard 1 and 2.

[Brezak]

The crash happens when we remove empty slabs in MeshAllocator::free_meshes. This may cause the slab ids in slab_layouts to no longer point to valid slabs. Instead of ignoring invalid slab ids in candidate_slabs could we remove them outright?

[eero-lehtinen]

That would probably be better.

Also does the number in SlabId matter? next_slab_id will be infinitely increasing over a long session of spawning and despawning things. If it's somehow directly mapped to memory we could run out.

[eero-lehtinen]

Now I made the (probably) proper fix.

[kristoff3r]

I agree that this is the right fix. It's true that free is now technically linear in the number of current slabs, but I don't think that will ever be a bottleneck. And it's simpler than breaking the invariant that all the SlabIds in the HashMaps are valid entries into slabs

[tychedelia]

This seems pretty straightforwardly correct compared to the previous fix.