[rlsw] Smarter texture management (#5296)

* simplified framebuffer management

* convert texture to rgba32

* remove float copy/blit

* adding SIMD framebuffer read/write
This adds SIMD framebuffer read/write, and also texel fetch.
This supports SSE2/SSE41 and SISD fallback (also includes ARM NEON support, conceptually identical but still needs testing).

* consistency

* tweaks

* review of `sw_texture_sample_linear`

* better quad sorting
unrelated to the PR, but at least it's done

* ignore some pipeline state in certains context

* convention tweaks

---------

Co-authored-by: Ray <raysan5@gmail.com>

Author: Bigfoot71