Retry all non-200 responses for send_message

[tadman]

Timeouts and other transient failures should be treated the same way as a concrete error message from the API.

Queue and retry according to the usual mechanism.