Report git pull conflicts as sync errors instead of false success

[Copilot]

Manual repository sync could report success even when git pull aborted due to local uncommitted changes or merge conflicts. That left sync_status green, cleared sync_error, and advanced the sync timestamp although the working tree was not updated.

• Sync task outcome handling

  • Teach sync_repo() to treat sync_repository() results prefixed with error: as real pull failures.
  • Persist sync_status = "error" and the returned git message instead of overwriting the repo row with success.
  • Skip last_synced_at updates on failed pulls so the repo state continues to reflect the last successful sync.

• Regression coverage

  • Add a task-level test that reproduces the dirty-working-tree case against a real bare remote:
    • remote advances
    • local clone has uncommitted edits
    • git pull aborts with “would be overwritten by merge”
    • repo row remains in error state and local edits are preserved

• Behavioral impact

  • The UI polling path now sees the existing sync_status === "error" / sync_error fields for failed pulls, so sync failures surface correctly instead of showing a false green check.

result = sync_repository(repo.local_path, repo.default_branch)

if result.startswith("error:"):
    detail = result.removeprefix("error:").strip() or result
    repo.sync_status = "error"
    repo.sync_error = detail[:500]
    session.commit()
    return {"status": "error", "message": detail}

repo.last_synced_at = datetime.now(UTC)
repo.sync_status = "success"
repo.sync_error = None