Prune unreachable bootstrap nodes

Also fix a bug affecting gateway connections. Previously, the gateway would try to ping the remote address during the connection protocol. The dial timeout is 2 minutes, so the peer can appear unresponsive for up to 2 minutes before RPCs start succeeding. This seems to confuse muxado, leading to dropped connections. When I scanned the bootstrap list, only pre-v1.0.3 peers were consistently reachable. (The ping was introduced in v1.0.3.) However, it is also clearly possible to successfully connect to (and accept connections from) nodes of any version. So it's unclear what exactly causes the failure.