Additional cleanup and simplification of hwintrinsic instruction sets for xarch

[tannergooding]

This does the remaining follow up work on #116230 to reduce the number of InstructionSet_ on the JIT side and make things a bit simpler to handle

[dotnet-policy-service]

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

[tannergooding]

@MihuBot

[Copilot]

Pull Request Overview

This pull request simplifies and cleans up the handling of hardware intrinsic instruction sets on the JIT side by replacing several dependencies on SSE41 (and related sets like BMI1/BMI2/FMA) with newer ones (SSE42, AVX2, AVX512v2, etc.). The key changes include:

• Updating intrinsic dependency checks and reset IDs from SSE41 to SSE42 throughout jit's lowering, codegen, and other components.
• Replacing BMI1/BMI2/FMA references with AVX2 in several cases.
• Removing the EnableIncompleteISAClass configuration from jitconfigvalues.h.

Reviewed Changes

Copilot reviewed 35 out of 35 changed files in this pull request and generated no comments.

File	Description
src/coreclr/jit/lsrabuild.cpp	Updated intrinsic dependency from InstructionSet_SSE41 to InstructionSet_SSE42.
src/coreclr/jit/importercalls.cpp	Swapped FMA and POPCNT dependency checks to use AVX2/SSE42 intrinsics.
src/coreclr/jit/jitconfigvalues.h	Removed the EnableIncompleteISAClass configuration setting.
(Other files)	Similar updates across lowering, codegen, hwintrinsic*, emitter, etc.

[tannergooding]

CC. @dotnet/jit-contrib, @EgorBo

Last of the JIT side cleanup related to the ISA simplification work. Was able to remove another ~1200 lines of code.

[EgorBo]

LGTM, but I am not sure I understand why various Insert/Extract/Round instructions were moved under SSE42, could you explain?

[tannergooding]

LGTM, but I am not sure I understand why various Insert/Extract/Round instructions were moved under SSE42, could you explain?

With the simplification, the JIT doesn't actually support targeting just SSE3, SSSE3, or SSE4.1 anymore. So the JIT is able to be simplified by combining all of these into SSE4.2. In some cases this allows fewer code paths to exist, because we no longer need to have a separate path to support SSE4.1 vs SSE3. It also makes it clearer that no JIT path needs to consider these differences, reduces the number of cases that switches need to handle, etc.

The downside is that the JIT InstructionSet no longer match the managed class names 1-to-1. But the throughput and other simplifications we get are worth it overall.