*: test watch restore in network-partitioned node, clean up fields, logging

[gyuho]

To cover the real-world use case that fails 100% without @yudai's watch fix #9281.

diff --git a/mvcc/watchable_store.go b/mvcc/watchable_store.go
index edbb7f9d0..bdffbf749 100644
--- a/mvcc/watchable_store.go
+++ b/mvcc/watchable_store.go
@@ -193,7 +193,7 @@ func (s *watchableStore) Restore(b backend.Backend) error {
        }
 
        for wa := range s.synced.watchers {
-               s.unsynced.watchers.add(wa)
+               s.unsynced.add(wa)
        }
        s.synced = newWatcherGroup()
        return nil

To make it testable, added SnapshotCountCatchUpEntries field, but make it configurable only via tests.

[codecov-io]

Codecov Report

❗ No coverage uploaded for pull request base (master@1c904b1). Click here to learn what that means.
The diff coverage is 95.12%.

@@            Coverage Diff            @@
##             master    #9745   +/-   ##
=========================================
  Coverage          ?   68.26%           
=========================================
  Files             ?      374           
  Lines             ?    35104           
  Branches          ?        0           
=========================================
  Hits              ?    23964           
  Misses            ?     9366           
  Partials          ?     1774

Impacted Files	Coverage Δ
etcdserver/raft.go	79.8% <ø> (ø)
embed/etcd.go	67.66% <100%> (ø)
etcdmain/config.go	82.8% <100%> (ø)
embed/config.go	58.91% <100%> (ø)
integration/cluster.go	83.38% <100%> (ø)
etcdserver/config.go	84.45% <100%> (ø)
etcdserver/server.go	72.76% <92.3%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 1c904b1...31094e5. Read the comment docs.