…st (#21321)

# Objective

It should be possible to enable minimal bevy features by disabling all
and then progressively enabling them until everything works.
This however requires, that bevy has good error reporting and gracefully
supports different featuresets.

I ran my code with minimal features and got this unhelpful error:
```
thread 'main' (993068) panicked at /home/jakob/dev/rust/bevy/crates/bevy_render/src/render_graph/graph.rs:158:26:
InvalidNode(PostProcessing)
```

## Solution

With this PR it looks like this:
```
thread 'main' (989393) panicked at /home/jakob/dev/rust/bevy/crates/bevy_pbr/src/wireframe.rs:136:14:
node PostProcessing does not exist
```

Which immediately helps me figure out that I need some feature for the
`WireframePlugin` I added.

Co-authored-by: Alice Cecile <[email protected]>

# Objective

#21297 removed the restriction
that prevented macos/ios from using clustered decals

## Solution

Remove them from the list of "unsupported" targets.

…incorrect pass ordering (#21348)

Light tiles used to be generated _after_ the world cache update, despite
the world cache update relying on them. This means that the world cache
update used last frame's light tiles, which is fine for static lights,
but completely wrong for dynamic lights and lead to missing GI
contributions from dynamic lights.

Moving the presample light tile step to before the world cache update
fixes this.

Can be tested by running the solari example, turning off the directional
light so there's only the emissive robot light, enabling
VISUALIZE_WORLD_CACHE, and then comparing before/after this PR.

# Objective

Fixes #21319.

## Solution

In the "about_to_wait" state of Bevy's winit event loop, we borrow
WINIT_WINDOWS only to check the conditions for a redraw, and perform the
call after that check is concluded.

## Testing

Tested by confirming that the RefCell error no longer occurs in my
application. In my application, the error occurred when the
`bevy-egui-inspector` crate was used.

I do not know whether the original reporter of #21319 was also using
third-party Bevy plugins that might have caused the issue. It would be
good if they could check whether this change fixes it for them.

Tested on Windows 11, where the bug is known to occur. The code that was
changed is only included in Windows builds so there is little need to
test on other platforms.

---------

Co-authored-by: Alice Cecile <[email protected]>
Co-authored-by: Dimitrios Loukadakis <[email protected]>

# Objective

Fixes #21293
Fixes #17314 to ensure that this is tested correctly.

## Solution
when despawning an entity, previously the swapped in (archetype) or
moved in entity (table) (which both require extra bookkeeping to update
archetype or table rows) were marked as `spawned_or_despawned` by the
location and tick that the to-be-removed entity was meant to be marked,
while the to-be-removed entity wasn't marked.
As pointed out by @akimakinai in
[#19047](#19047), I've re-added
the correct `mark_spawn_despawn` call to `despawn_with_caller`.

## Testing

I've added a test `spawned_after_swap_remove` that ensures that
despawning an entity by swapping doesn't effect other entities
`spawned_or_despawned` location, and that it does effect the despawned
entity's index's `spawned_or_despawned` location.

Co-Authored By: [email protected]

---------

Co-authored-by: WaterWhisperer <[email protected]>