[rtexture] Cubemap mipmap loading improvements

[Not-Nik]

Another two patches for loading cubemaps with mipmaps. Hopefully the last ones 😅

Two fixes in one PR for loading cubemaps with mipmaps:

1. Previously we were always generating mipmaps if we could. This makes sense for cubemaps with some mipmaps, but not all possible. If we didn't do it there, that cubemap would be incomplete when filtering with GL_LINEAR_MIPMAP_LINEAR/RL_TEXTURE_FILTER_MIP_LINEAR or GL_NEAREST_MIPMAP_LINEAR/RL_TEXTURE_FILTER_NEAREST_MIP_LINEAR, because those require all possible mipmaps to be available¹. For cubemaps with no mipmaps (just the base mipmap level), there is no need to generate any mipmaps (or use VRAM to store them), because the user wouldn't expect there to be any mipmaps in the first place. Thus, only build cubemap mipmaps when necessary.
2. Previously, a cubemaps mipmap count would always be set to 1 (on the CPU), regardless of the actual number of mipmaps present on the GPU. This change neatly ties in with the first patch, because we can just use the number of mipmaps the image has, were we may or may not have generated them.

I tested that the model_skybox example still works, and obviously I tested it with my own IBL setup, where everything also works.

As last time, I structured the PR to be able to just be rebased onto master if you want all the different changes to reflect in the tree.

Footnotes

1. This is not entirely true for "normal" OpenGL, but since there is no way to set GL_TEXTURE_MAX_LEVEL in plain raylib, I'd say it applies for us. See "The base and max levels must only specify mipmap levels that have been allocated [...]" here ↩

[raysan5]

@Not-Nik Nice improvement! Thanks!