Retry on errors from Dynamo about cortex

[jml]

@juliusv in #84 points out that we do retry in normal operation but not during shutdown.

[jml]

Fixed by #119

[tomwilkie]

Still not doing retries on shutdown

[jml]

Ah right. Kind of curious now as to what @juliusv meant when he said we were backing off already.

[tomwilkie]

retry, not backoff.

its because we don't drop the chunks if we fail to persist them, so they'll get retried on a later pass.

[tomwilkie]

#153 makes us retry and throttling for dynamodb, even on flush.

We still don't retry other errors, but we haven't observed any yet.