[Feature needed] parquet-format reader for HDFS

## Background

We have notice that if we only use raw angel (not spark-on-angel), we can only choose to read a single text file as input. If the input data source is large (> 1G), we may need to store the data as parquet format in HDFS. Maybe it's a good idea to implement a parquet-format reader for raw angel.

## Suggestion:

add a option `parquet` for param `angel.input.format` , and implement it in the same directory of BalanceInputFormat.java