libhdfs++: potential segfault after teardown

During shutdown, it is possible for an RPC task to outlive the RPC connection and segfault when it attempts to signal the engine on timeout.

In rpc_connection.h, the callback

req->timer().async_wait(std::bind(
      &RpcConnection::HandleRpcTimeout, this, req, _1));

does not retain a reference to shared_this, so the connection can be destroyed.