agent architecture - recovery from agent shutdown

[aviadr1]

if an agent (implemented by users of livekit using the python sdk) crashes, then ideally the livekit-backend should continue pinging the agent pool until another agent becomes available.

however this is not the case: if the agent stops (for instance I am running the agent in my debugger, and stop the debugger to change code), and then run the agent again, the livekit-backend never calls my agent to rejoin the room.

for a system built with agents to become production-ready, this error recovery mechanism is a must