[mlir] Fold trivial subtensor / subtensor_insert ops.

Static subtensor / subtensor_insert of the same size as the source / destination tensor and root @[0..0] with strides [1..1] are folded away.

Differential revision: https://reviews.llvm.org/D96991