naga: fix panic parsing SPIR-V subgroup reduce/scan/ballot ops

[nazar-pc]

Connections

Fixes #8389.

Description

Disclaimer: this PR was mostly done using a clanker and I do not fully understand the code changes, but I have reasonably high confidence in their correctness.

Several OpGroupNonUniform* handlers in the SPIR-V frontend called emitter.finish() before reading their argument operand via get_expr_handle!. When the argument's defining expression came from a control-flow body that had already closed (e.g. a value computed inside a loop and used after it), get_expr_handle needs to spill the value through a temporary local variable, which requires calling emitter.finish() itself. Since the emitter had already been finished, this second call unwrapped a None and panicked.

Move the emitter.finish() call to after the argument expression handle(s) are obtained for OpGroupNonUniformBallot, the arithmetic/logical reduce-scan ops, the broadcast/shuffle ops, and OpGroupNonUniformQuadSwap.

Testing

Existing CI tests pass, new regression test was added and original SPIR-V file from #8389 no longer crashes naga CLI. The project where I originally hit this issue also passes all tests with this fix applied (Linux, AMDGPU, Vulkan).

Checklist

• I self-reviewed this PR.
• I fully understand this PR.
• WebGPU implementations built with wgpu may be affected behaviorally.
• Validation and feature gates are in place to confine behavioral changes.
• Tests demonstrate the validation and altered logic works.
• CHANGELOG.md entries for the user-facing effects of this change are present.
• The PR is minimal, and doesn't make sense to land as multiple PRs.
• Commits are logically scoped and individually reviewable.
• The PR description has enough context to understand the motivation and solution implemented.

[jimblandy]

This looks fine, but please trim the snapshot output languages.

[jimblandy]

Thank you!