nir: a few 16-bit fixes
Here's some patches factored out from !10101 (merged). I believe also !10185 (merged) would benefit from the first two patches.
The first two patches is about splitting nir_shader_compiler_options::support_8bit_alu
into a separate flag for integer and floats.
The last patch fixes an issue I triggered when trying to run do yuv to rgb conversion on a texture lookup that was lowered to 16 bits.
/cc @pendingchaos for the first two patches.