Previously _load_registry silently dropped dead PIDs, so jobs that
completed in the background were invisible on restart. Now dead PIDs
are loaded as 'success' status so the user sees they completed. The
registry is cleaned up after loading.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Jobs are now saved to gniza-jobs.json in WORK_DIR when they start and
finish. On TUI restart, the registry is loaded and PIDs are checked —
still-running jobs appear in the Running Tasks screen and can be
killed. Reconnected jobs are polled every second to detect completion.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Jobs were tied to screen @work tasks, so switch_screen cancelled the
worker and lost the process reference. Now start_job uses
asyncio.create_task so jobs survive screen changes and can be killed.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
SIGTERM was being ignored by child processes. Use SIGKILL via the
actual process group ID, with proc.kill() as fallback.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Start CLI subprocesses in their own session so SIGTERM via
os.killpg reaches child processes (rsync, etc.) not just the shell.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Backup and restore operations now run as background jobs instead of
blocking modal screens. Users can navigate away and check progress
from a dedicated Running Tasks screen. OperationLog supports attaching
to running jobs with live output polling.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
