Replace subgroup query operations in TritonGEN with their GPU dialect equivalent

[FMarno]

Subgroup query operations were added to the GPU-To-LLVM-SPV lowerings recently, so now the gpu dialect operations can be used instead of TritonGEN SubgroupIdOp and SubgroupLocalIdOp.

When this change is present in the target LLVM checkout, we should remove TritonGEN SubgroupIdOp and SubgroupLocalIdOp and use GPU SubgroupIdOp and LaneIdOp directly.