Implement backoff for transport reconnection

[jdknives]

Feature description

Currently, when one edge of a transport goes down, the other edge will try to re-establish the transport every three seconds. We should implement a backoff for re-establishing the transport. @evanlinjin suggested using the retry logic in /internal/netutil . Also, we need to improve logging, so that these messages do not clutter stdout:

[2019-11-13T10:59:10+08:00] WARN [tp:03c775]: failed to redial underlying connection: dial tcp 192.168.0.107:7777: connect: connection refused

• improve logging
• implement retrial backoff