JIT/wasm: Pri-1 R2R codegen fixes (BBJ_CALLFINALLY back-edge, null-check loads)

[AndyAyersMS]

Two wasm JIT codegen correctness fixes surfaced by the Pri-1 R2R test pass.

BBJ_CALLFINALLY non-fallthrough back-edge
genCodeForBlock early-returns for BBJ_CALLFINALLYRET assuming fall-through to
the next linear block. On wasm the finally is invoked via call_indirect, so when
the paired continuation is reached via a back-edge (e.g. a catch leave targeting a
loop header outside the enclosing try/finally) no branch was emitted and control
fell out of the wasm loop, running a cloned finally a second time. Emit a br to
the continuation when it is not the next linear block. Fixes the double-finally in
strswitch_catchrettoinnertry and similar.

Null-check faulting GT_IND loads
On wasm a load from a null address does not fault, so a decoupled array bounds
check read the length from [null+lenOffset] as 0 and threw
IndexOutOfRangeException instead of NullReferenceException. genCodeForStoreInd
already emits an explicit null check for faulting stores; do the same for faulting
GT_IND loads (lower marks the address multiply-used, regalloc reserves the temp,
codegen emits the check using the same GTF_EXCEPT && OperMayThrow predicate the
throw-helper reservation phase uses). Fixes b42929, b41063, b31903,
b091942, and the JitTest_catchfinally_* array-null tests.

[AndyAyersMS]

@adamperlin PTAL
fyi @dotnet/wasm-contrib

[Copilot]

Pull request overview

This PR updates the CoreCLR WASM JIT backend to address two correctness issues that arise from WASM’s different control-flow and faulting semantics compared to native targets (notably: call_indirect finally invocation and non-faulting loads from address 0).

Changes:

• Ensure BBJ_CALLFINALLY codegen emits an explicit branch to the BBJ_CALLFINALLYRET continuation when that continuation is not the next block in the linear (preorder) block order.
• Add explicit null-check emission for faulting GT_IND loads in WASM codegen, mirroring existing explicit null-check handling for faulting stores.
• Plumb the required “multiply-used address” temp consumption through lowering and WASM regalloc so the address is available both for the null check and the subsequent load.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
src/coreclr/jit/regallocwasm.cpp	Consumes the temporary register for multiply-used GT_IND addresses so the null-check+load sequence can share the address value.
src/coreclr/jit/lowerwasm.cpp	Marks faulting GT_IND load addresses as multiply-used so codegen can both null-check and load using the same computed address.
src/coreclr/jit/codegenwasm.cpp	Emits explicit null checks for faulting loads and emits a corrective branch after call_indirect finally calls when the continuation is non-fallthrough.

[dotnet-policy-service]

Tagging subscribers to 'arch-wasm': @lewing, @pavelsavara
See info in area-owners.md if you want to be subscribed.

[AndyAyersMS]

@adamperlin ping

[adamperlin]

Taking a look!

[adamperlin]

Overall lgtm, I just had two questions.

[AndyAyersMS]

@adamperlin@SingleAccretion revised, ptal