#242 BUG: Workflow retry leaves completed_at set causing corrupted state

When a workflow fails and Absurd retries it, _update_workflow_run_if_exists_first_start() sets status='running' but does NOT clear completed_at. This causes corrupted state where status=running AND completed_at is set. Root cause: orchestrator.py line 1033 Fix: Added completed_at = NULL and error = NULL to the UPDATE statement. Symptoms: - Workflow shows 'running' but has completed_at timestamp - Cancel API fails because state is inconsistent - Workflows loop indefinitely (fail, retry, fail, retry...) Fixed in this commit.