Ignore fallthrough blocks in optRecognizeAndOptimizeSwitchJumps

[EgorBo]

Fixes a codegen issue @stephentoub noticed in #107499

There were 2 issues:

1. Phase ordering - "optimize bools" phase glues BBJ_COND(return true/return false) back to BBJ_RETURN, so switch recognition should be earlier (and it used to be earlier)
2. Empty BBJ_ALWAYS fallthrough blocks confuse the optimizer - I didn't want to call fgUpdateFlowGraph just because of them so I applied a fix here.
3. I enabled the switch-recognition for ARM64, just bumped the number of checks (driven by diffs). Still had to keep it disabled for ARM32

Diffs

[Copilot]

Pull Request Overview

This PR fixes a codegen issue in the JIT's switch recognition optimization by addressing two problems: phase ordering and handling of empty fallthrough blocks. The optimization was running too late in the compilation pipeline, after the "optimize bools" phase had already converted conditional jumps back to returns, making switch recognition ineffective.

Key changes:

• Moved switch recognition phase to run before boolean optimization phase
• Added logic to skip empty BBJ_ALWAYS fallthrough blocks during switch pattern matching

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
src/coreclr/jit/switchrecognition.cpp	Added SkipFallthroughBlocks helper function and updated target resolution to handle empty fallthrough blocks
src/coreclr/jit/compiler.cpp	Moved PHASE_SWITCH_RECOGNITION to run before PHASE_OPTIMIZE_BOOLS

[dotnet-policy-service]

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

[EgorBo]

PTAL @dotnet/jit-contrib fixes an optimization that someone broke (see PR desc).

[amanasifkhalid]

fixes an optimization that someone broke (see PR desc).

That was probably me. Thanks for fixing this!

[EgorBo]

@AndyAyersMS I've addressed your feedback, looks like Aman's approval is no longer green, so can you approve? thanks