render: add color transform option for render pass output

This is an attempt to rebase and continue !4090 (closed) . I am posting this since adding support for output color profiles is a lot of work, and having a partially working design may help either myself or others design a more polished MR.

To test this, use the corresponding Sway PR, https://github.com/swaywm/sway/pull/7681 .

This PR is mainly concerned with getting the necessary infrastructure to color manage outputs. There are a few possible improvements which I probably will not have the time to do myself in the next few months.

• Color transform precision is low -- a 33x33x33 3D LUT can't properly capture e.g. the linear segment of the linear->sRGB function. To implement the ICC lutBtoAType, would need a multistage transform ( Matrix→3x(1d LUT)→Matrix->3x(1d lut)→3d LUT->3x(1d lut) .) [This might be worth delaying to a future MR, as it would probably cost >1000 LoC to handle all the cases.]

• That color transforms actually interpret the ICC profiles correctly, and white point correctly, etc.

• The wlr_color_transform and its addons (wlr_vk_color_transform) currently are kept alive as long as there are things that require either; it might be more efficient to delete the wlr_color_transform as soon as there are no references to it, and then keep the wlr_vk_color_transform alive until all rendering involving it has completed. This probably isn't critical.

• When using a non-8-bpc output buffer format, taking a (shm-type) screenshot may redraw pixels without the color transform applied. types/wlr_screencopy_v1.c#L211-216 uses the old renderer API, and in particular calling wlr_renderer_end() will run the second subpass that copies the pixels from in the blend_image to the main buffer, without using a color transform. !4211 (merged) would probably fix this, as it no longer calls wlr_renderer_begin()/wlr_renderer_end() in the screencopy code. (Fixed and merged in separate MR).

• Need to block direct scan-out when applying a nontrivial color transform. (Fixed on Sway impl)

• Validation errors when not using color management: need to bind a dummy lut_3d texture to the output pipeline. (done)

• Flickering -- some frames render with a garbage color transform, suggesting a race condition. Does anyone know what the issue is? (Fixed, was a buffer span offset/alignment issue)

• Requires that the blend_image is already set up; i.e., the output transforms are only enabled when non-8-bit buffer formats are in use. (Fixed, currently ugly solution)

• Creating a 3d lut texture on every frame is inefficient; typical use is unlikely to change the output ICC profile rapidly

• Should keep wlr_vk_color_transform alive until all commands involving it complete.