fix(codegen): keep captured loop counters on shared storage

[proggeramlug]

Summary

• keep captured or otherwise non-local loop counters out of the static i32 local-bound specialization
• make loop conditions and updates continue sharing the counter's heap box
• add an end-to-end regression with a timeout for both unused and retained closure captures

Root cause

The static counter < localBound optimizer created a fresh stack i32 shadow even when the counter was a captured mutable var. The loop update correctly wrote the shared heap box and returned before synchronizing that shadow, while the optimized condition kept reading the shadow's initial value. The condition therefore stayed true forever.

Tests

• cargo test -p perry --test local_bound_loop_semantics -- --nocapture
• cargo test -p perry-codegen --lib
• cargo fmt --all -- --check
• scripts/check_file_size.sh
• git diff --check

Closes #11052

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

Note

Currently processing new changes in this PR. This may take a few minutes, please wait...

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 5a0c279d-9674-474d-8a99-956a6956ffb4

📥 Commits

Reviewing files that changed from the base of the PR and between f5cfbff and 23c4dd2.

📒 Files selected for processing (2)

• crates/perry-codegen/src/stmt/loops.rs
• crates/perry/tests/local_bound_loop_semantics.rs

 ___________________________________________________________________
< Your commit message says 'quick fix'. That's how folklore starts. >
 -------------------------------------------------------------------
  \
   \   \
        \ /\
        ( )
      .( o ).

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[proggeramlug]

Landed on main in merge train 259 (#11079), released as v0.5.1642 at 7f4417b5a1.

Your commits are on main — cherry-picked from this PR's head a7c8de9fce and validated as one tree, CI 22/22 green including all 6 gap-suite shards. GitHub cannot auto-close a PR whose commits land under new SHAs via a train rebase, which is why I'm closing this by hand rather than it happening on merge.

Nothing is lost and nothing is needed from you. Thanks.