Resolve one compositing space per camera stack

[stuartparmenter]

Objective

Part of the HDR upstreaming work. The display pipeline coming in later PRs needs cameras that share a render target to agree on one compositing space, since they composite into one shared buffer and later passes have to know how it's encoded. Two follow-up PRs build directly on this, the camera-stack composition contract and the shared writer-encode path.

Getting there also fixes real bugs that exist today. Stacked cameras can request different CompositingSpaces, and each keys its own pipelines, so they write differently encoded pixels into one buffer and silently composite garbage. Oklab on a texture format that can't store its negative values breaks just as silently, as do requests on render paths that never encode.

Solution

A new bevy_render::view::composition module resolves every view's request once per frame into a ResolvedCompositingSpace component. A stack resolves to the single distinct encoded request among its members, or to linear when there is none. Misconfigured shapes resolve to linear with a warning. Solo views and non-stack groups keep their own requests.

Every consumer reads the resolved value instead of the raw request, including the sprite and mesh2d keys, the clear-color conversion, and the upscaling blit key. ViewTarget::compositing_space is removed, with a migration guide entry. Two 2D specialization systems move later in the schedule so their caches see this frame's resolved value.

Testing

Added lots of new tests, Check, clippy, and tests pass.

---

This PR was built by me with the assistance of Claude Code w/ Fable 5

[JMS55]

Review of things minus composition.rs (haven't gotten to it yet).

Idea is sound to me, mostly some nits and some things I'm worried about changing.

[JMS55]

Reviewed composition.rs now. Overall idea is fine, but the comments could use a lot of work. They're making things more confusing, not less.