Recover from socket-based UDF connection failure?

It looks like kapacitor does not attempt to reconnect to a socket-based UDF if the connection fails for any reason. Worse, this results in task stopping:

`[stream_reduc_avail_heaters:TestUDF2] 2017/02/06 13:45:40 E! stopping UDF server: write error: write unix @->/tmp/test-udf: write: broken pipe
[task_master:main] 2017/02/06 13:45:40 E! Stopped task: test_task TestUDF2: stopping UDF server: write error: write unix @->/tmp/test-udf: write: broken pipe`

Recovery apparently requires disabling and re-enabling the task, or restarting the kapacitor daemon. I'm using `socat` to forward from a UDS to a (potentially non-local) TCP socket, so connection failure is a real concern. What is the suggested way of dealing with that? I was hoping there would be a configuration option to automatically restart the task, perhaps with an exponential back off.

Apologies if I missed something obvious in the docs.

No files, tests, or entry points are named. Start by locating the socket-based UDF connection and task-stopping paths, then review how failures are propagated. Done would require an agreed recovery behavior for failed connections, including whether and how task restart or backoff should work.

Written by the indexing model from the issue text.