Optimize meshlet material pipeline preparation

[CrazyRoka]

Objective

• Reduce CPU overhead in the meshlet renderer by avoiding redundant computations.
• Optimize prepare_material_meshlet_meshes_main_opaque_pass and prepare_material_meshlet_meshes_prepass performance during the render preparation phase.

Solution

The previous implementation performed expensive computations for every meshlet instance before checking if the resulting pipeline_id was already in the cache.

I refactored the caching logic to check the cache beforehand. The heavy computation is now wrapped inside the cache-miss block.

Testing

• I used a separate stress test (introduced in Add many_meshlet_materials stress test example #23792) to measure the impact of these changes.
• The issue happens when unique-materials flag is enabled: cargo run --features=meshlet,https --release --example many_meshlet_materials -- --unique-materials
• Also compared 3d/meshlet example behavior before and after the fix.

---

Showcase

Before - 52% CPU usage (Flamegraph)

After - 6.7% CPU usage (Flamegraph)

[JMS55]

Thanks for the PR!

Noting for future reference: The real fix will be to make this retained on each entity, similiar to how the standard Bevy renderer was restructured a few releases ago.