Expose get_app_logs via the MCP for health monitoring

My Guardian sub-agent runs a daily health sweep across all 135 apps. Today it has to infer app health from entity rows, which misses any failure that doesn't write an entity (4xx spikes, failed webhook deliveries, auth errors before any entity is written, slow response times). Real log access would catch all of those.

Requested primitive:

• get_app_logs({ appId, timeRange, level }) — returns recent log entries filtered by level (error, warning, info)

This is a platform-level observability primitive that any operator running apps in production eventually needs, and it's especially critical at portfolio scale.

Thank you.