fix: wait for actual Claude response

[JamesW546]

Summary

• ignore hook and system lifecycle events instead of treating them as successful responses
• require a real AssistantMessage followed by a successful ResultMessage
• isolate the wake-up request from user/project hooks, MCP servers, and CLAUDE.md settings

Problem

The previous string-based check treated HookEventMessage as success because its class name does not contain SystemMessage, even though it is a system lifecycle event. The function then returned before Claude produced an assistant response, causing false-positive wake-up logs.

Verification

• python -m py_compile waker.py
• manual wake-up request completed with AssistantMessage
• scheduled cron run completed with a real user -> assistant session