Support backoff if rate limited

[stevejalim]

While we're not likely to hit it (famous last words), there is a rate limit on the Smartling API, so we should build in a backoff mechanism so we don't end up in a broken state.

Whether that can be an in-process backoff or moving to a worker queue approach is up for discussion

[stevejalim]

The simplest implementation here would be a an in-process backoff - basically a gently increasing time.sleep(), but wondering what sensible backoff steps would be.

Smartling's API docs don't appear to disclose what the limit is - perhaps that depends on the customer or plan. However they do recommend an in-process time.sleep approach in their docs, plus say:

A typical value for the maximum delay is 30-60 seconds; and for the maximum number of retries, 5-10.