Improve palette handling in STB backend

[ankith26]

fixes libsdl-org/sdl2-compat#492

While I was investigating that issue I realised that this must actually be fixed on the SDL_image end and not sdl2-compat end as I initially assumed.

This PR makes the stbimage backend path more like the libpng backend:

• If a colorkey is set along with a palette, the colorkey entry shouldn't have redundant pixel alpha of 0 as that messes up the SDL_MapRGB logic.
• The stbimage backend always gives 256 entries in the palette, even if less entries are required. It seems like stb image does not expose this information in its API. I have put a bit of a hacky solution to this issue. It's certainly not very efficient but is the easiest approach I could think of.

[ankith26]

If this PR is accepted I can remake this PR against main/SDL3

[slouken]

Iterating over all the pixels to find the highest index is relatively expensive. Is there a change to stb_image that we can make, to retain the correct palette size?

[ankith26]

Is there a change to stb_image that we can make, to retain the correct palette size?

I'm not aware how to do this rn. My assumption here is that its very unlikely to have an 8 bit image that is also very large, so I think the added complexity should be okay-ish.

[slouken]

Merged, thanks!