Improve codegen for ExtractMostSignificantBits

[tannergooding]

No description provided.

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[tannergooding]

CC. @dotnet/jit-contrib

Simple improvement to ExtractMostSignificantBits to use the EVEX mask support for short/ushort. This gives some significant codegen improvements, FullOpts (-10,667 bytes), including to known impactful SIMD functions supporting char. There are no throughput differences.

The diffs typically look like:

; Vector128 Example, saves 3 bytes
-       vpshufb  xmm0, xmm9, xmmword ptr [reloc @RWD00]
-       vpmovmskb ecx, xmm0
+       vpmovw2m k1, xmm9
+       kmovb    ecx, k1

; Vector256 Example, saves 9 bytes
-       vpshufb  ymm0, ymm0, ymmword ptr [reloc @RWD00]
-       vpermq   ymm0, ymm0, -40
-       vpmovmskb eax, xmm0
+       vpmovw2m k1, ymm0
+       kmovw    eax, k1

Additionally it adds the minimal support to DecomposeLongs so that Vector512.ExtractMostSignificantBits can be used on x86 (32-bit).

[tannergooding]

This should be ready for review now