fix: restore signal handlers on lock release

[davidberenstein1957]

Closes #1310

What changed

codecarbon/lock.py:

• Lock.__init__ now keeps the handlers returned by signal.signal() in self._previous_handlers instead of discarding them.
• _handle_exit releases the lock and then delegates to the handler it replaced: it calls a previous Python handler, or reproduces the default disposition with os.kill(os.getpid(), signum) when it was SIG_DFL, or does nothing for SIG_IGN. It no longer raises SystemExit(1) unconditionally.
• release() restores the saved handlers (only when the currently installed handler is still ours, so an application that registered its own afterwards is not clobbered) and unregisters the atexit hook.
• The atexit callback is stored once as self._atexit_hook, because atexit.unregister() will not match a freshly-created bound method.

Why

Lock is constructed whenever allow_multiple_runs=False. Because it overwrote the process signal disposition permanently, restoring the default SIGINT handler never happened and KeyboardInterrupt stopped being raised at all — every except KeyboardInterrupt: in the embedding application became dead code, including CodeCarbons own in codecarbon/cli/monitor.py:95. Applications that had registered a graceful-shutdown SIGTERM handler lost it silently.

Verification

Two new tests in tests/test_lock.py (TestLockSignalHandlers):

• test_release_restores_previous_handlers — a sentinel SIGTERM handler is installed, a Lock is built, and the sentinel (plus the original SIGINT handler) must be back after release().
• test_signal_is_forwarded_to_previous_handler — raises a real SIGTERM and asserts the applications handler ran.

Both fail on master and pass with this change; uv run pytest tests/test_lock.py -q is green (7 passed). uv run task format/lint reformat the whole repository with the current tool versions, so only the two touched files were formatted and checked (ruff/black clean apart from pre-existing findings in tests/test_lock.py).

Ordering note

The stop() idempotency fix should land first. Now that _handle_exit chains to the previous handler, the CLIs own SIGINT handler (codecarbon/cli/main.py) runs after the lock releases and calls tracker.stop(); on a session that already stopped the tracker that becomes a second stop(), which would turn a hijacked signal into a duplicate emissions row.

Behaviour change

Previously any SIGINT/SIGTERM ended in SystemExit(1). Now the process does whatever the application asked for. That is the intent of the fix, but anyone relying on CodeCarbon force-exiting on Ctrl-C will notice.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.51%. Comparing base (065d0e6) to head (a91020f).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1329      +/-   ##
==========================================
+ Coverage   91.39%   91.51%   +0.11%     
==========================================
  Files          49       49              
  Lines        5056     5067      +11     
==========================================
+ Hits         4621     4637      +16     
+ Misses        435      430       -5     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.