Prefer agent CLI for continual learning hook

[ericzakariasson]

Summary

• prefer launching the agent CLI from the continual-learning stop hook when agent is available on PATH
• fall back to the existing followup_message behavior when agent is unavailable or cannot be started
• document the new hook behavior and best-effort hook log file in the plugin README

Testing

• executed the hook logic through a Node TypeScript shim to validate the non-triggering path
• simulated a triggering turn without agent on PATH and confirmed the hook still emits the original followup_message
• simulated a triggering turn with a stub agent on PATH and confirmed the hook returns {} and launches agent -p --force --workspace ...

Slack Thread