Help with another push inconsistency edge case (#2131432)

* If we are forced to abort a "ping" due to a watchdog alarm, we should
  handle this as a ping failure (which potentially changes the heartbeat)
  rather than a garden variety IOException.
* This prevents the additional overhead of connection error backoffs,
  which would only tend to increase the time needed to recover from the
  error.
* In one case reported by Moto, this appears to be the behavior of a WiFi
  router with NAT timeout.  This fix will cause maximum delay for pushed
  mail to be reduced in most cases.

Change-Id: I2b0e3b10d82762d20f63cac3ac4638a03f13f842
1 file changed