Need to add log or finer metrics about duration

[lance6716]

Feature request

currently during locating the cause of synchronization lag, we found some information is missing:

1. for current blocking checkpoint flushing, we need to know the duration of "preparing checkpoint flushing" which is equal to the time DM stops sending job to synchronizing queue.
2. for current replication lag updating frequence, it is much more frequently updated than metrics, so metrics is not reliable. We might print the lag or count when lag is higher than a threshold, or using histogram metrics.