NIR: Misc 64-bit lowering and vec8 fixes

Faith Ekstrand requested to merge gfxstrand/mesa:review/nir-64-vec8-fixes into master

The int64 lowering pass bug I found while enabling load/store vectorization. The others I found while looking at a failing test.

cc: @mattst88 and @karolherbst as potentially interested parties

Merge request reports