JIT: refine SSA info in ambiguously threaded blocks

[AndyAyersMS]

If we partially jump-thread a block with PHIs, the remaining preds that target block may all bring in the same SSA def for some of the PHIs. If so, remove the PHIs and update the SSA/VN for the PHI uses in the block.

In particular there may just be one ambiguous pred.

Fixes #126703.

[AndyAyersMS]

@EgorBo PTAL
fyi @dotnet/jit-contrib

diffs [updated]

[dotnet-policy-service]

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

[Copilot]

Pull request overview

Refines CoreCLR JIT optimizations around jump-threading through PHI-containing blocks by improving SSA/VN rewriting when only ambiguously-threaded predecessors remain, and adds a small assertion propagation enhancement to fold comparisons against negative constants when the compared value is proven non-negative.

Changes:

• Track and remove redundant PHI-def statements in partially jump-threaded blocks when remaining incoming paths agree on the same SSA def, and rewrite SSA/VN for affected uses.
• Adjust successor SSA replacement computation to reflect post-threading reaching paths.
• Enhance assertion propagation to fold ==/!= against negative constants when the value is known non-negative; add a regression test with disasm CHECKs.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
src/tests/JIT/opt/RedundantBranch/JumpThreadPhi.cs	Adds disasm CHECK expectations to validate improved codegen for the repro pattern.
src/coreclr/jit/redundantbranchopts.cpp	Implements PHI-use rewriting + optional PHI removal when ambiguous preds converge on a single SSA def after partial threading.
src/coreclr/jit/assertionprop.cpp	Adds folding for X ==/!= (negative const) when X is proven never-negative.

[AndyAyersMS]

From the inspiring example

	Control flow and IR before RBO. We are going to jump thread through BB4, but it's only partial threading since we don't know how BB4's predicate evaluates when control comes from BB3.
	Partial jump threading is done. This is how things are left in main. But with this PR we now realize that there is just one reaching def for V3 in BB4, and so we rewrite the use and remove the PHI
	Assertion prop is then able to propagate the "never negative" range and eliminate the compare in BB4.
	Post-AP there is just one compare left...