Expose near lossless WebP encoding

[DeeJayLSP]

Bugsquad edit removed, as it is now invalid.
I still intend to use this PR for the implementation, so I ask to keep it open.

Adds a checkable option to texture importers called Near Lossless, when Mode is set to Lossy:

It also affects other Lossy importers:

When checked, importing will use the same workflow as Lossless, but the value of Lossy Quality will be passed to config.near_lossless, which decreases image quality in order to reduce its size (while keeping RGB(A) color space instead of using lossy's YUV(A)420) when below 100. This is a built-in feature in libwebp.

A few things to note:

• WebP Compression Level project setting will affect images imported as Near Lossless. Overhaul WebP packer and split compression options #67948 kinda solves this.
• The image will be imported as true lossless if Lossy Quality is set to 1.0 (config.near_lossless' default is 100) or the image being imported is too simple.

[Calinou]

Out of curiosity, how does Near Lossless differ from Lossless Lossy quality 100 in practice? This should be documented in the class reference.

[DeeJayLSP]

Out of curiosity, how does Near Lossless differ from Lossless quality 100 in practice? This should be documented in the class reference.

It doesn't. Near Lossless at 100 is true lossless. Maybe you meant Lossy quality 100?

Lossy encodes images using YUV420, while Near Lossless and Lossless use RGB. That's the main difference.

[Calinou]

I meant Lossy indeed.

Lossy encodes images using yuv420/yuva420p, while Near Lossless and Lossless use RGB/RGBA. That's the main difference.

I see, that makes sense. I've used lossy quality 100 for godot-docs images that were too large to use lossless compression, and it still looks pretty good while having file sizes smaller than JPEGs with lower quality.

[DeeJayLSP]

Despite the fact it works, I didn't like how messy the code came out. May do a complete refactor later by using helper functions that will likely reduce diff.

[DeeJayLSP]

Refactor done.

The main functionality of the webp lossless function was turned into a helper function, while the original calls this helper by providing a near lossless value of 1.0 (actually lossless).

The lossy function, that previously called the modified lossless function, now calls the helper.

Everything that was changed to match the new lossless function was reverted.

[fire]

Looks good.

As per custom:

• needs a contributor review
• needs style and architecture review

[mhilbrunner]

The feature is IMO nice, the code looks good and I spotted no style issues or problems and is fairly straightforward. I left a few nitpicks as a suggestion for the documentation above. Thanks for contributing!

[DeeJayLSP]

I have changed the helper function a bit to better match my follow-up PR.

WebPConfig.quality is a float value, but WebPConfig.near_lossless is an integer.

[DeeJayLSP]

I'm not thinking about closing this in favor of #67948 as its parts are more obvious, and also sort of fixes compression level (which is now replaced) affecting near lossless directly.

But changes done there may not be approved like this one.

This PR will be on wait for #67948 (and maybe some changes by reduz) to be done, as it's better to be done on top of it.

stale, OP has further plans for PR