Improve State restart time on failure caused by server restart, by using activity heartbeat, when startToClose is large

[longquanzheng]

Currently when iwf server restarts, the state api will fail and wait for next attempt by the startToClose timeout + backoff retry interval.
If the startToClose timeout is very large (e.g. >10 mins), it will wait for a long time. To avoid the unnecessary waiting, Temporal/Cadence has a concept of "activity heartbeat" to tell Temporal/Cadence server that the worker is still alive. If no heartbeat is received within heartbeat timeout, Temporal/Cadence will reschedule next activity immediately based on backoff retry policy.

Note: this is also because of the fact that Temporal/Cadence activity task/worker is "polling based". iWF task/worker is "pushing" so it doesn't have such issues.