More work on handling GuaranteedExecutor contexts when shutting down NetworkInterfaceTL. Previous work blocks shutdown until they've all executed, and uses an existing cancel mechanism for the top-level CommandState/ExhaustCommandState promise/future pairs.

What remains is a need to cancel the chain of futures hung off the future returned from AsycDBClient::runCommandRequest(). The good news here is that we can return a network error, due to an onError() added in RequestState::resolve() prior to scheduling the remaining work on a GuaranteedExecutor. We just need to be able to guarantee we're not adding new work to the connection pool as we try to drain it.