#15619: fix: clean up orphan LaunchAgent plist on bootstrap failure

## Summary Fixes #14315 - When `launchctl bootstrap` fails during onboarding (e.g., user denies macOS security prompt), the plist file was left on disk with no cleanup path - Now attempts `launchctl bootout` and removes the plist file before throwing the error - Both cleanup operations silently catch errors (file may not exist, service may not be loaded) ## Changes - `src/daemon/launchd.ts`: 3 lines added — bootout + unlink on bootstrap failure - `src/daemon/launchd.test.ts`: New test verifying plist cleanup on bootstrap failure ## Test plan - [x] New test: "cleans up plist file when bootstrap fails" — creates failing launchctl stub, verifies plist is removed - [x] All 15 launchd tests pass - [x] No existing test assertions changed 🤖 Generated with [Claude Code](https://claude.com/claude-code) <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR updates the macOS LaunchAgent install flow to clean up on `launchctl bootstrap` failure: it attempts a `launchctl bootout` and unlinks the generated plist before rethrowing the bootstrap error. It also adds a regression test that stubs `launchctl` to fail specifically on `bootstrap` and asserts the plist is removed afterward. The change is localized to `src/daemon/launchd.ts`’s `installLaunchAgent` path and extends the existing launchctl-stubbing test harness in `src/daemon/launchd.test.ts`. <h3>Confidence Score: 4/5</h3> - Mostly safe to merge once the Windows test hermeticity issue is addressed. - Core runtime change is small and well-scoped, but the new test likely won’t correctly stub `launchctl` on Windows due to writing an extensionless batch file while execution relies on `shell: true` PATH resolution. Also noted a misleading dead `.catch()` in the cleanup path. - src/daemon/launchd.test.ts (win32 stub), src/daemon/launchd.ts (bootstrap failure cleanup) _{Last reviewed commit: 20dabad} <!-- greptile_other_comments_section --> <!-- /greptile_comment -->