Fix flapping tests

[talbright]

For server cluster setup fix, some ideas were drawn from samuel#76 (However, that solution was not complete and required additional refactoring.)

[talbright]

Known flapping tests:

+
+--- FAIL: TestNoQuorum (1.47s)
+       cluster_test.go:143: Disconnect from wrong server: expected=127.0.0.1:12431, actual=127.0.0.1:12428
+

+--- FAIL: TestDNSHostProviderReconnect (2.13s)
+       dnshostprovider_test.go:130: Connected to "localhost:12728". Finding test server index…
+       dnshostprovider_test.go:134: …trying "localhost:12728"
+       dnshostprovider_test.go:137: …found at index 0
+       dnshostprovider_test.go:151: Create returned error: zk: connection closed
+

+
+--- FAIL: TestFLWSrvr (0.00s)
+       flw_test.go:86: listen tcp 127.0.0.1:2181: bind: address already in use
+

[talbright]

Ran the following locally, no errors:

touch test.log
for i in {1..20}
do
    echo "running test $i"
    go test -v -race -coverprofile=profile.cov ./zk >& test${i}.log
  if [ "$?" -ne 0 ]; then
        echo "test $i failed"
    fi
    echo "test $i finished"
done