Add timeout and "clean" failure to HTTP API.

[jpittis]

#159 describes a bug where deleting a toxic through the HTTP API permanently blocks until TCP timeout. As suggested in the issue, we should have a "timeout and recovery of some sort" that returns a nice error code to the user explaining that they should file a bug report.

While we want to stop the API from ever hanging, I'm sure there'll be more edge cases in the future. Error codes are much nicer to look at than nothing.

cc @xthexder @sirupsen @archelan