Don't flock snapshot files

[serathius]

Ref #17194

Flocking to snapshot files was added unintentionally in #1927.
Even at this time PurgeFile function was used for both wal and snapshot files

etcd/etcdserver/server.go

Lines 323 to 339 in 502396e

	func (s *EtcdServer) purgeFile() {
	var serrc, werrc <-chan error
	if s.cfg.MaxSnapFiles > 0 {
	serrc = fileutil.PurgeFile(s.cfg.SnapDir(), "snap", s.cfg.MaxSnapFiles, purgeFileInterval, s.done)
	}
	if s.cfg.MaxWALFiles > 0 {
	werrc = fileutil.PurgeFile(s.cfg.WALDir(), "wal", s.cfg.MaxWALFiles, purgeFileInterval, s.done)
	}
	select {
	case e := <-werrc:
	log.Fatalf("etcdserver: failed to purge wal file %v", e)
	case e := <-serrc:
	log.Fatalf("etcdserver: failed to purge snap file %v", e)
	case <-s.done:
	return
	}
	}

Did minimal changes to avoid flocking files. Don't want to refactor it now to allow easy backport to older versions.

cc @ahrtr

[fuweid]

Not sure I understand correctly, if that server recovered from snapshot db, server will flock it by bbolt. If purge goroutine doesn't try lock, it will delete using data.

UPDATE: Before open backend, the file will be renamed to db.

[serathius]

/retest

[serathius]

ping @ahrtr