Spark streaming terminated when close meta data log error

Now, HDFSMetaDataLog is the only implement of MetaDataLog. If HDFSMetaDataLog.writeBatchToFile() execute error, then total app will be terminated.

The meta data log is very important for spark, we should retry some times when exception, and add a config to control retry times.

For example HDFS has a bug that close error if has no enough replicas detail HDFS-11486.