Add support for lz4 in parquet

Hadoop uses a native block format for LZ4 (same as parquet-mr api) which is incompatible with LZ4 block format.

As a result Parquet/LZ4 could have different block formats.

The parquet-cpp api (now Apache Arrow) uses LZ4 frame format, which is also incompatible with LZ4 block format.

The current decision is to use a format compatible with Hive, Spark and parquet-mr.