Propagate the traced script's sys.exit code without losing trace output

[AnshBajpai05]

Last of the small fixes from my fork — after this one I'd like to open a Discussion about upstreaming some larger features.

Symptom: any traced script that ends with sys.exit() — which includes most CLI tools and every argparse program — silently loses all oracletrace output:

$ oracletrace mycli.py --json trace.json
$ echo $?
0        # script's real exit code also swallowed if the trace aborted mid-way
$ ls trace.json
ls: cannot access 'trace.json': No such file or directory

runpy.run_path() lets the script's SystemExit fly straight through _trace_script, so the summary, --json/--csv/--html exports, and --compare never run, and oracletrace exits with whatever code the script raised — without having done its job.

Fix: catch SystemExit around run_path, keep tracing/reporting/exports intact, and surface the script's own exit code at the very end (Python semantics preserved: sys.exit() → 0, sys.exit(7) → 7, sys.exit("msg") → message on stderr, code 1). Gate codes (--fail-on-regression) still take precedence, since that's the CI contract.

Tests: four e2e subprocess tests — code propagated, exports written after sys.exit, string-argument semantics, bare sys.exit() success.

🤖 Generated with Claude Code

[KaykCaputo]

Nice, I hadn't thought about that. That's a very important point. Thanks! Feel free to continue the discussion.