fix cache refresh and add metrics

[keepsimple1]

1. The main purpose is to fix a bug in "cache refresh": the daemon would repeatedly send out query packets to refresh its cache when a record refresh_due returns True before the record expires.

Because there is a "20% of TTL" time span between "refresh" and "expire", this can cause an outgoing packets surge due to cache refresh. The fix is: only refresh once for any refresh-due record.

2. To debug this issue and to help similar debugging in the future, I also added a metrics (counters) for outgoing traffic. A user can easily pull the metrics by calling get_metrics of the daemon, and use recv() or try_recv().

3. Fix a problem in retransmission: currently the retransmissions is a BTreeMap and there is a non-zero chance that two retransmissions are scheduled with the same next_time, hence the 2nd one will overwrite the first one in the BTreeMap due to the same key. The fix is to change the data structure to a Vec.

4. Updated the integration test to cover the metrics.