perf(workflow): batch step persistence after execution

[xiami762]

Summary

• make workflow step callbacks pure in-memory collectors with no SQLite calls or event-loop waits
• persist all steps, the final execution summary, workflow stats, and retention changes in one atomic completion transaction
• use a dedicated completion connection plus a completion lock to prevent transaction interleaving
• skip the initial queued-row write for Kafka, Syslog, Poller, and custom high-frequency triggers while preserving complete step history at the end
• keep the manual run queued record, ToolContext lifecycle, cancellation, backpressure, restart, and no-overlap behavior

Performance validation

Real SQLite benchmark with 4 concurrent executions and 7 steps each:

• callback SQL calls: 0
• callback time per worker: 0.015-0.037 ms
• completion wall time for all 4 executions: 1.363 ms
• completion commits: 4
• persisted steps: 28
• persisted executions: 4
• workflow stats call count: 4

Test coverage

• targeted workflow persistence suite: 85 passed
• broader workflow and ingest regression suite: 374 passed, 2 skipped
• 2 unrelated NDR workflow fixture failures remain on the broader unfiltered suite
• Ruff and git diff checks pass

Persistence guarantees

• step callbacks never call record_step, record_steps, or run_coroutine_threadsafe
• each completed execution commits once
• failures rollback steps, execution summary, stats, and retention together
• executionLog remains empty in the summary row; step details remain queryable from the step table