Improve sync thread alerting mechanism

* When a sync thread triggers an alarm by failing to return from
  an HttpPost beyond the socket timeout, we call abort() on the
  HttpPost to force it to stop
* It appears that there are cases in which this is insufficient,
  and the thread remains hung in a blocked state
* The result of this failure is to prevent the syncing mailbox from
  ever syncing again, and is typically seen by a failure to receive
  new mail (as reported in the referenced bug)
* In this CL, we add code to wait for 10 seconds after calling the
  abort() method.  If the HttpPost is still hung, we interrupt() the
  thread, and have SyncManager release the Mailbox, so that another
  thread can be started.

Bug: 2615293
Change-Id: I6a48195fc68bb950126006326a5b30448d3bbb63
3 files changed