web_platform: copyExternalImageToTexture with more source types (PNG, ImageData) with color spaces

[kainino0x]

Followup to #913.

The PNG tests in particular are important because they should have wider browser support than ImageData's colorSpace option.

Here are some test images: https://bugs.chromium.org/p/dawn/issues/detail?id=1139#c8 I took the WebKit test image and cropped it to 1x1 in GIMP. Source: https://webkit.org/blog-files/color-gamut/

• red1x1-displayp3.png: https://webkit.org/blog-files/color-gamut/Webkit-logo-P3.png cropped to 1x1 and saved with the same color profile.
• red1x1-untagged.png: https://webkit.org/blog-files/color-gamut/Webkit-logo-P3.png cropped to 1x1 and saved with no color profile.
• red1x1-srgb.png: https://webkit.org/blog-files/color-gamut/Webkit-logo-sRGB.png cropped to 1x1 and saved with the same color profile.

Verified they display as expected on my MacBook (displayp3 is brighter than the other two).

We can test all three of these. The displayp3 red should be > 1, the others should be == 1.

• [ ] Wide color gamut images
  • See above
• [ ] High bit depth images
  • 16 bit PNG (Safari technically might support TIFF so we could test that too if we want)
  • Precision should be preserved all the way into the texture (rgba32float or rgba16unorm texture for example)
  • Can probably just copy the WebGL test, which is very simple: https://github.com/KhronosGroup/WebGL/pull/3578 That was generated using this script: https://github.com/kainino0x/wide-color-test-images
• [ ] HDR images
  • There are various encoding (icc profile, hlg/pq, jpg/avif, non-gainmap/gainmap, etc.?)
  • TBD what formats/encodings should be tested, TBD what the results should be
  • See also #3193

[kainino0x]

Edited summary above to add that we need to test high bit depth images as well.

[kainino0x]

Here's a python script for generating test images. It has a few basic examples in it. https://github.com/kainino0x/wide-color-test-images

[kainino0x]

Also added HDR to the list above.

The test case I have in mind is a simple case similar to the 10bpc one I added to WebGL, but with 16 red pixels:

[[   0    1    2    3]
 [1020 1021 1022 1023]
 [1024 1025 1026 1027]
 [and 4 values leading up to the max brightness of the encoding]]
where SDR 1.0 = 1023

with added checks to make sure that the values are in the right range (not just that there are enough distinct values, like in the 10bpc test).