Ensure Vector.Sum uses SSE3, rather than SSSE3, for floating-point

[tannergooding]

This resolves #54119 and was introduced in #53527

CC. @zlatanov, @echesakovMSFT

[tannergooding]

This wasn't caught by the ISA tests that were run as those only run for the runtime side tests and only library side tests were added.

I've added a runtime side test, mirroring the other runtime tests for Vector, to help ensure this gets the per-ISA validation.

[tannergooding]

/azp run runtime-coreclr jitstress-isas-x86, runtime-coreclr jitstress-isas-arm

[azure-pipelines]

Azure Pipelines successfully started running 2 pipeline(s).

[zlatanov]

Thanks @tannergooding. Looking at this now I can see that SSSE3 HorizontalAdd is not supported for floats and doubles, but I must have missed it when using it.

I actually tried to see if somehow the macros that are generating the NI_ enums could also have comments generated with them, which would include information for which native types the instruction is supported, but I failed.

For example I think it would be invaluable to have something like:

/// <summary>Supported for ...</summary>
NI_SSSE3_HorizontalAdd

[tannergooding]

For example I think it would be invaluable to have something like:

That information is all in the metadata table here (either an instruction or INS_invalid, there are a few cases where invalid can still generate something, but those are rare and typically "helper" intrinsics): https://github.com/dotnet/runtime/blob/main/src/coreclr/jit/hwintrinsiclistxarch.h#L348

The expansion of a macro isn't normally visible without explicitly running an explicit compilation step and saving the pre-processor output, so I don't think having a comment would be beneficial.

[tannergooding]

/azp run runtime-coreclr jitstress-isas-x86, runtime-coreclr jitstress-isas-arm

[azure-pipelines]

Azure Pipelines successfully started running 2 pipeline(s).

[tannergooding]

ARM64 failures are unrelated (#54125).