sh_lut fixes and changes

The notable addition is that we now support linear sampling emulation for arrays/literals, which turns out is actually a performance gain for the no-compute shader polar sampling fallback code. (This effectively moves the LUT access out of the texture fetch unit thus allowing it to avoid bottlenecking the texel fetches for the texture itself)

Needs testing to make sure it isn't a horrific performance regression on some older iGPU or something