Add ICC profile to ImageDecoder

[sophie-h]

I license past and future contributions under the dual MIT/Apache-2.0 license,
allowing licensees to choose either at their option.

---

Giving this a go. Not sure why &mut self was suggested in the issue

Closes #1832

[fintelia]

Thanks for this PR! The reason to require &mut self is because some future decoders might require it. In particular, we don't want to force every decoder to parse the ICC profile in its new() method.

[sophie-h]

Changed all to &mut

[HeroicKatora]

👍 Though I now loathe how ImageDecoder::read_image takes self instead of &mut self. For instance, in the PNG decoder the ICC profile may not be part of the header but be placed at some arbitrary location in the data stream. The caller can not generically know if the information has already been read until the image parsing was 'done'. The whole design deserves a good rework.

Going to ignore the deny warnings for this PR, it looks like rav1e and criterion had new releases we should bump to.

[sophie-h]

in the PNG decoder the ICC profile may not be part of the header

I guess I should mention in the docs that it could only be available after completely reading the image?

The whole design deserves a good rework.

Not sure if that is about this MR? I mean, that sounds like a fundamental issue. But for ICC profiles, it seems like it's fine to only have them after the complete image is decoded.

For EXIF, it would be much more relevant to know it as early as possible because knowing the size and orientation of the image is very helpful for the presentation, even before complete decoding.

[HeroicKatora]

Not sure if that is about this MR? I mean, that sounds like a fundamental issue. But for ICC profiles, it seems like it's fine to only have them after the complete image is decoded.

It's not, PR itself looks good to me.