Distributed builds uses too much traffic to determine agent state

Currently the code performs a ping before and after every operation to check that the agent is operating correctly.

It would be better if it could periodically check if agents are up (CONTINUUM-2615 suggests that the agent could report in), and enable/disable accordingly.

For normal operations, a ping should not be needed on either side - instead fail normally if the request fails, and if it is an operation that can move to a different agent, call again after disabling.