Fix Switch swizzle texture conversion and optimize Texture2DConverter

[isHarryh]

Summary

This PR mainly fixes the issue of Switch texture decoding, including:

• Fixes Alpha8 swizzle texture decoding and encoding.
• Fixes RGB24 swizzle texture decoding and encoding.
• Fixes DXT5 swizzle texture decoding and encoding.
• Fixes incorrect swizzle detection (platform blob).

This PR also does some improvements and small fixes to the Texture2DConverter:

• Removes the unnecessary copy procedure in parse_image_data, making it ~5% faster.
• Removes the unnecessary paste procedure in pad_image, making it ~20% faster.
• Fixes an error in swap_bytes_for_xbox that bytes object is not mutable.
• Changes A mode to L mode in image_to_texture2d, because A is not a valid PIL image mode.
• Refactors the CONV_TABLE.
  • Now TextureFormat -> (converter function, (additional args, ...)).
  • Typo of EAC_R_SIGNED has been fixed.
  • ETC series uses "ETC2_RGB"-like fmt instead of [2, "RGB"]-like.
• Wraps the functions in the TextureSwizzler.

Related Issues

Fixes #317

Fixes #319

[isHarryh]

Tests

Samples provided by #317 and #319 .

Format	Decoding	Encoding
Alpha8 swizzle	√	√
RGBA32 swizzle	√	√
BGRA32 swizzle	?	?
RGB24 swizzle	√	√
BGR24 swizzle	?	?
DXT1 swizzle	?	?
DXT5 swizzle	√	√

√ = tested, ? = not tested but theoretically okay

tests_result.zip

[isHarryh]

Oh! It seems you forgot about my PR and independently resolved the swizzle texture issue in version 1.22.4.

However, there are still some issues with your implementation.

Due to commit conflicts, I’ll go ahead and close this PR and open new PRs to address these problems, while also including the other optimizations from this PR.

@K0lb3

[K0lb3]

I implemented a fixed version before I saw that you included it as well, otherwise I would have sticked and requested a fair amount of changes.
Imo PRs that fix things should more or less stay atomar and only implement the fix, and nothing else.
This makes it easier to review and therefore gets merged faster.

Regarding the changes I would have requested, it would have started with using the formatting defined by rust.toml.
I might be at fault for that not being correct, as I borked the inclusion pattern in ruff.toml, but in the end it could still be seen in the git difference.

Some asserts you included aren't really necessary, at least the type checker didn't complain to me about those points, and I don't really see how a type checker could get confused by some of them.

Anyway, in case I didn't annoy you too much with my behavior,
I would be happy to merge threeseparate PRs based on your changes.

• the alpha fix
• the changes to etc and eac
• the removal of the unnecessary paste

[isHarryh]

Sorry for changing too many things in one single PR. I am also happy to draft new separated PRs later to fix these remaining things.

[isHarryh]

Fine. Now all the changes this PR addressed has been implemented in 1.22.5.