GPU: Multiview support

[ada-tv]

• I confirm that I am the author of this code and release it to the SDL project under the zlib license. This contribution does not contain code from other sources, including code generated by a Large Language Model ("AI").

Description

Allows drawing to multiple slices of array textures with a single render pass, handy for single-pass stereo, cubemaps, and shadow cascades.

I don't actually have any experience using multiview in practice, I was hoping to learn it using the SDL RHI lol

Any pointers from senior devs with more experience are appreciated

• Integration into testgpu_spinning_cube_xr

API stuff

• Should BeginMultiViewGPURenderPass take an integer view count (Metal) or a bitmask (Vulkan and D3D)? Assuming view count is the common denominator, going with that
• View offsets?
• Behavior of BeginMultiViewGPURenderPass when only given one view
  Just render to the first array slice?
• Multiview state might need to be set up in graphics pipelines

Backends

• Vulkan
  • Should KHR_multiview be enabled by default or should we require it to be enabled manually through a property?
    Will expose real max view count if multiview is available, falls back to 1 if not
• Direct3D 12 (don't have a Windows machine to test with)
• Metal (don't have a macOS box to test with)

Existing Issue(s)

Closes #16215

[flibitijibibo]

Should KHR_multiview be enabled by default or should we require it to be enabled manually through a property?

If the XR properties are present we can make multiview mandatory, otherwise we can ignore/skip it.

[ada-tv]

If the XR properties are present we can make multiview mandatory, otherwise we can ignore/skip it.

For now I'm making it so the max view count is exposed through a device property defaulting to 1, so if an app wants to fall back to another method or quit with an error message it can do so.

Some drivers on Linux either don't support multiview or have only started supporting it relatively recently.

[flibitijibibo]

Latest version of the support query bits lgtm. Once the spinning cube example is updated I'll test on-device ASAP.