Improved error handling of Pollers

[laksnv]

Excluding handling of Throttling exceptions, Flux currently doesn't have a mechanism to backoff before retrying Poll requests. There are at least a couple of cases where it makes sense to back off by different amounts:

• LimitExceededException - When service returns this, there are already close to 1000 Polls waiting in SWF for Tasks. Polling more is just waste of resource
• Internal Errors - When service is experiencing issues, I think it makes sense to back off a bit before retrying

There is a fixed 10ms backoff between Poll requests, which can be improved.

Thoughts?

[danielgmyers]

I'll need to look at the content of the SdkServiceException class to see if there's some convenient helpers for this, but handling additional cases of server-side issues like this as part of the catch (SdkServiceException e) block makes sense.