tests: normalize LLVM IR when updating goldens

[jakebailey]

This bugged me while working on another branch, as -update kept overwriting all of my golden files with LLVM 22 outputs instead of the normalized ones. I think it's better to just store it normalized.

[Copilot]

Pull request overview

This PR updates the LLVM IR golden-file workflow so regenerated goldens are stored in a normalized form (to reduce LLVM-version churn) and updates an existing golden accordingly.

Changes:

• Update -update paths in transform/compiler tests to avoid unnecessary golden rewrites when outputs are equivalent across LLVM versions.
• Normalize LLVM-version-specific IR spellings before writing goldens.
• Refresh compiler/testdata/large.ll to the normalized (pre-LLVM21-style) attribute spellings.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
transform/transform_test.go	Adjusts golden update logic for transform tests and applies IR normalization when writing .out.ll files.
compiler/compiler_test.go	Adjusts golden update logic for compiler tests and normalizes IR when writing .ll goldens; also refactors comparison to reuse actual.
compiler/testdata/large.ll	Updates a golden file to use normalized IR spellings (e.g., nocapture instead of captures(none)).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[dgryski]

LGTM. At some point we should update the version of LLVM that we're normalizing to.

[jakebailey]

The repo's llvm-project clone is 22 so arguably that should be the one being used...

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated no new comments.

[dgryski]

The various regexps in normalizeIR rewrite to pre-LLVM22 versions though (captures(none) -> nocapture, etc)

[jakebailey]

Right, I didn't change the status quo in this PR, I was just suggesting 22 for some unspecified future date (which could be now if desired?)