Solari: Prevent world cache cells from keeping each other alive infinitely

[JMS55]

Objective

• Prevent world cache cells from querying each other and keeping each other alive infinitely
• Now when you leave an area of your game, you're no longer paying performance cost for the out-of-view cells.
• Also fix a long-standing off-by-1 bug in the world cache compaction code

Solution

• Instead of always resetting cell lifetimes to the max lifetime during a cell query, for world cache update, we now set it to the max of the current cell lifetime and it's existing lifetime (max prevents a lower-lifetime cell overwriting a higher-lifetime cell).
• Credit to IsaacSM and @NthTensor for the idea!

Testing

Video showing the current number of live world cache cells, before and after this PR.

Screen.Recording.2025-11-21.111911.mp4

[NthTensor]

Side note, we need to figure out why the bounce lighting isn't super visible in this test scene. Maybe @DGriffin91 knows?

[JMS55]

Unfortunately this is a bit more expensive, likely because of the extra atomics. I'll have to test if copying the lifetime before the cell update to avoid the atomicLoad() improves things.

[JMS55]

Side note, we need to figure out why the bounce lighting isn't super visible in this test scene.

        DirectionalLight {
            illuminance: light_consts::lux::FULL_DAYLIGHT,
            shadows_enabled: false, // Solari replaces shadow mapping
            ..default()
        },

With default exposure.

[NthTensor]

Unless we can find actual visual issues with using a store instead of a max, the store is probably acceptable. It just means we will be a bit more aggressive at clearing than is really necessary.

[JMS55]

atomicMax vs atomicStore doesn't seem to matter. I'm guessing the new atomicLoad is the problem, which is easy enough to fix.

[DGriffin91]

Side note, we need to figure out why the bounce lighting isn't super visible in this test scene. Maybe @DGriffin91 knows?

Which version of the scene is this? By default all the materials are 100% metallic.

Side note: I'm not able to open the scene in bevy 0.17: #21490

[JMS55]

Figured out the performance changes, it's just from adjusting the maximum allowed cell lifetime. Not a big deal.

[JMS55]

Which version of the scene is this?

@SparkyPotato's on github. Your's wasn't updated at the time I set this up.