A refresh can jump across multiple threads, we cannot just lock a mutex, need to track who that is and clear when finished.
In addition to that, fix our internal syncrepl session tracking pulling
it closer to RFC 4533,
refreshDone now represents the receipt of
refreshDone flag. Refreshing status is maintained for plain refreshes
(and times when we might be starting one but don't know for sure).
We still reschedule a new sync with a failure delay if there is another one running but tracking refreshes this way paves the way to being able to wake them up if we start tracking them somehow.