[LMA] Pacemaker fails to restart the collector when the watchdog check fails

Pacemaker monitors /tmp/lma_collector.watchdog to verify that the Heka pipeline isn't blocked.

Unfortunately the implementation is buggy on at least 2 points:
- Pacemaker doesn't stop the running Heka process when it detects that the watchdog file is too old.
- The /tmp/lma_collector.watchdog file isn't removed when the Heka process is stopped. So Pacemaker may consider that the new Heka process is down relying on the old watchdog file.