Improve GitHub API fetch reliability

[jbmorley]

There are a few simplifications and improvements that can be made to the overall GitHub fetch infrastructure; GitHub supports paging at the HTTP layer using the Link header which should allow for a common implementation of paging as an asynchronous iterator across all APIs.

A common fetch implementation should also support back-off and retries to ensure we're robust to GitHub's rate limits. (Unfortunately, there seems to be no getting around needing to do multiple fetches unless we implement a service end-point which is notified of changes, mantains state, and provides its own HTTP API.)