Bevy 0.19 VRAM usage regression on Vulkan

## Bevy 0.19

System: 7950x, 128GB ram, 4070ti, 16GB vram.

If I run [bistro](https://github.com/bevyengine/bevy/tree/main/examples/large_scenes/bistro) or [sponza](https://github.com/DGriffin91/bevy_sponza_scene) using the Vulkan backend (on both linux and windows) I OOM:
`ERROR bevy_render::error_handler: Caught rendering error: Out of Memory`

This doesn't appear to be an issue with the Metal or DX12 backends. If I set `WGPU_BACKEND=dx12` on windows I am able to open multiple instances of bistro without issues, and can see that I am spilling into system RAM as expected. This was also the behavior with the Vulkan backend in Bevy 0.18

Tried bisecting, ran into different issues/crashes at different times. I believe the issue started when Bevy updated to wgpu 28 and the issue persists with wgpu 29.

@JMS55 Started work on a wgpu PR to try to resolve the issue: https://github.com/gfx-rs/wgpu/pull/9462

### Timelime
Before 543b305adce77f84cffaa964b631724ea7051135 the scene loads as expected. Each of these are the first commit that the behavior changes:

- 1/22: 543b305adce77f84cffaa964b631724ea7051135 crashes with `Trying to destroy a SwapchainAcquireSemaphore that is still in use by a SurfaceTexture`
- 1/25: d734e7feb060900f4a79f5611126c272e271e31e crashes with `wgpu error: Out of Memory`
- 2/5: 71dd9ea7db45b9d32f38d08c8abd45e98b1fdc31 doesn't crash but scene is flashing
- 5/6: 4bd3708d523b6ec00e40c798c7f4e77dbd1c1788 crashes with `wgpu error: Out of Memory`

Previously in the 0.19 dev cycle, the oom would result in cause flashing, this was resolved. This older issue addressed both the flashing and the oom: https://github.com/bevyengine/bevy/issues/23183

Reproduce the issue with the bistro or sponza scene using Vulkan, then compare the behavior with DX12 or Metal. Review the listed regression commits, the older issue #23183, and wgpu PR #9462; done means Vulkan scenes load without the out-of-memory error or flashing.

Written by the indexing model from the issue text.