fix: read by direction when resampling the prefiltered radiance highest level

[sasvdw]

PR Details

Summary

Mip 0 of the prefiltered cubemap lost continuity around every face edge whenever the copy shortcut did not
apply, because the fallback rescaled each face on its own.

Description

RadiancePrefilteringGGXNoCompute fills mip 0 from the source when DoNotFilterHighestLevel is set. The
copy at :103 serves that only while the sizes, the mip count and the format agree. Everything else went
through ImageScaler over a single face view, which clamps at the face border, so the border texels never
saw the neighboring face.

This replaces that fallback with CubemapFaceResampleShader, which reads the source by direction. A
border texel then crosses the face edge the way the hardware samples a cubemap.

Falling through to the existing filter would look like the tidier fix, and it is what the compute path
appears to do, but it does not work. At roughness 0 the GGX distribution returns a zero probability
density, the solid angle goes to infinity, and the sample mip clamps to MipmapCount
(RadiancePrefilteringGGXNoComputeShader.sdsl:41-44). Every texel reads the smallest mipmap and the level
becomes one flat color. That is why mip 0 carries a special case at all.

Measured over the 12 shared face edges, against the step between neighboring texels inside a face:

	within-face	seam ratio
before, ImageScaler	4.03	0.86
falling through to the filter	0.00	0.00
after, resample by direction	5.16	0.11

The middle row is the trap. A flat level reports no seam, so a continuity check alone would call it fixed.
The test asserts the level still carries an image for that reason.

Motivation and context — see #3334.

Related Issue

Fixes #3334.

Types of changes

• Docs change / refactoring / dependency upgrade
• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Checklist

• My change requires a change to the documentation.
• I have added tests to cover my changes.
• All new and existing tests passed.
• I have built and run the editor to try this change out.

Validation status

• Tests: TestRadiancePrefilteringGgxFaceContinuity drives the filter with an output format that differs
  from the source, which is the case SkyboxGenerator creates for a low dynamic range skybox. It measures
  the step across the 12 shared face edges and asserts the level still carries an image. Red at 0.86 before
  the change, green at 0.11 after, verified both ways.
• No gold image moves. Every cubemap in the tree takes the copy path, so nothing that is captured today
  reaches the changed branch. Stride.Graphics.Tests.11_0 passes 5 of 5, including
  TestRadiancePrefilteringGgx. The skybox cases in Stride.Graphics.Tests.10_0 pass 3 of 3.
• Not covered: the same roughness 0 problem sits in the compute path if its size check ever fails, and
  that path also skips the format check the no-compute path makes. Both are described in the issue and left
  alone here.
• Editor: not exercised.

[Ethereal77]

Why is this skipped for Vulkan?

[sasvdw]

It was a bad copy on my end. Thought I was still working with the code from another branch where I had the fixes living, for skipping some related tests and ended up duplicating the test suite without a proper look.