[X86] WIP Optimize SMULO/UMULO for AVX512 mask registers.

[craig.topper]

Summary

This further improves the codegen from D97624.

I'm considering merging this code into the new helper function
from D97624. So we would have Ovf outparam instead of Low. This
would avoid needing to reverse the TRUNCATE since we could
intercept the vXi16 result directly.

[craig.topper]

Merged into D97624