Split ordered and unordered TypeId maps

[0xEgao]

Objective

Fixes #24642.

TypeIdMap became backed by IndexMap in #23864, making order-preserving removals O(n), including removal-heavy component registration paths.

Solution

• Restored TypeIdMap as a hash-backed map with average O(1) removal.
• Added TypeIdIndexMap for call sites that require insertion order.
• Kept ordered storage for gizmo handles, custom attributes, and related-scene processing.
• Replace unnecessary shift_remove calls with remove.

No new tests were added. Existing affected tests and compilation checks pass.

[beicause]

Kept ordered storage for gizmo handles, custom attributes, and related-scene processing.

From the comment GizmoHandles needs to preserve order. But I don't know the reason for the other two places.

[0xEgao]

Kept ordered storage for gizmo handles, custom attributes, and related-scene processing.

From the comment GizmoHandles needs to preserve order. But I don't know the reason for the other two places.

• CustomAttributes preserves insertion order because its existing iteration test explicitly relies on attributes being returned in builder insertion order.
• ResolvedScene::related is iterated while related entities are spawned and relationship targets are populated. Keeping it ordered makes entity allocation and relationship processing deterministic.

[alice-i-cecile]

I like this change, but I worry about the migration path. Can we do TypeIdIndexMap and TypeIdHashMap for explicitness, keeping TypeIdMap as a deprecated type alias for the former with a deprecation note about choosing between them?

[0xEgao]

I like this change, but I worry about the migration path. Can we do TypeIdIndexMap and TypeIdHashMap for explicitness, keeping TypeIdMap as a deprecated type alias for the former with a deprecation note about choosing between them?

Hmm makes sense. Changing TypeIdMap in place silently changes its iteration semantics and Entry API.
I’ll add explicit TypeIdIndexMap and TypeIdHashMap aliases, and keep TypeIdMap as a deprecated alias for TypeIdIndexMap.

[0xEgao]

I have made the changes, the changeset grew larger because all internal uses were migrated to explicit TypeIdHashMap or TypeIdIndexMap aliases. This avoids deprecated internal usage and demonstrates the intended migration path.

cc -: @alice-i-cecile