The EgressRequestRateLimiter doesn't wire up support for max queue depth because it didn't make sense to "reject" a request to write a response out, doing so would corrupt or disable the connection given the serial nature of mongorpc connections. This change covers enabling max queue depth support with a fail-open policy: when the queue depth is exceeded, just do the write immediately.  The finite depth gives us real protection by bounding the number of parked threads that can back up in the system, with an added benefit of using its overflow rate as a saturation signal for open-loop detection.