Make "remote:" resilient against connection failure

If a remote connection is dropped, it should be possible to re-establish the connection and resume the session, without the loss of any in-flight requests/contexts/services/etc. As a corollary, it might be worth exploring having more than one connection in a "bundle" to help parallelize the transit load and avoid head-of-line bottleneck problems. Perhaps it is worth looking into multihoming as well.

SSL contexts might be useful here.