nir: Trim unused components from nir store intrinsics
Following on from !6050 (merged), trim the unused write_masks and num_components of store intrinsics, producing a modest (.3%) instr win on i965/vec4. This is also part of making nir-to-tgsi awesome.