[mlir][sparse] disable nonunit stride dense vectorization

This is a temporary work-around to get our all-annotations-all-flags
stress testing effort run clean. In the long run, we want to provide
efficient implementations of strided loads and stores though

Reviewed By: bixia

Differential Revision: https://reviews.llvm.org/D98563