[AMDGPU] Skip unclusterd rescheduling w/o ld/st

We are attempting rescheduling without load store clustering
if occupancy limits were not met with clustering. Skip this
for regions which do not have any loads or stores at all.

In a set of kernels I am experimenting with this improves
scheduling time by ~30%.

Differential Revision: https://reviews.llvm.org/D97342