health checks

[Geal]

right now sozu evaluates the health of a backend server from TCP connections (if we cannot connect, we retry, then there's a circuit breaker after a while).

We should be able to measure the health from other means:

• timeouts
• HTTP 500 errors
• configurable heath checks on a status page configurable per application

that last one could be handled by the main process, and the server status dispatched to the worker. We can also send back an event to configuration clients to indicate the health check failed