fix: serialise _change_working_dir to stop cwd races (#3553)

[lsmlhi_25 (FOWEPJF255)]

Summary

• Serialise _change_working_dir, including creation of relative working directories, with a module-level threading.Lock so concurrent threads in the line-execution process pool cannot observe each other's intermediate os.chdir state.
• Add regression tests for basic restore, exception restore, mkdir, and thread-safety (fixes [BUG] Race condition with global state in process pool #3553).

Context

Reclaims the approach from stale closed PR #4205. Issue has 0 open competing PRs.

Test plan

• Targeted context-utils tests: 5 passed on Python 3.12.10.
• Black formatting, Python compilation, and git diff --check passed for the changed files.

Disclosure: drafted with AI assistance; personally reviewed against #3553 repro and prior #4205.

[github-actions]

Hi, thank you for your interest in helping to improve the prompt flow experience and for your contribution. We've noticed that there hasn't been recent engagement on this pull request. If this is still an active work stream, please let us know by pushing some changes or leaving a comment.

[lsmlhi_25 (FOWEPJF255)]

Yes, this PR is still active. I pushed a follow-up fix that moves relative directory creation under _working_dir_lock, so os.makedirs cannot resolve a path against another thread's temporary working directory. I also added a regression test that checks directory creation happens while the lock is held.

Validation: the targeted context-utils test file passes 5/5 on Python 3.12.10; Black formatting, Python compilation, and git diff --check pass. The latest PR checks report triage, CLA, and Snyk passing; no test workflow is currently reported.

Commit: 27dae5c

[lsmlhi_25 (FOWEPJF255)]

The latest parallel unit and end-to-end workflow runs both ended with action_required before executing any jobs (zero jobs and no logs): unit run and e2e run. So these runs have not tested this PR yet. Separately, the workflow files on main declare report jobs with needs: test, while the corresponding jobs are named parallel-unit-test and parallel-e2e-test; that appears to be a pre-existing workflow configuration issue. The focused local tests and checks for this change pass.