Activity dispatching optimization

[wxing1292]

copying issue from cadence: https://github.com/uber/cadence/issues/2965

[bergundy]

From the original issue:

before writing the remaining un-started activity (activity with retry policy only, since these are at least once execution) task to db, try pushing these task to matching service for sync match (e.g. wait for 1 to 2 second), if sync matching is a success, the corresponding transfer task can be omitted.

Leaving this here to be considered, the rest has already been implemented.