Support gorilla encoding for float numbers

[Gorilla](https://www.vldb.org/pvldb/vol8/p1816-teller.pdf) is a de facto encoding algorithm for float numbers, it has been used by many time series database such as InfluxDB, TimescaleDB for a while.

For now Parquet only supports BYTE_STREAM_SPLIT encoding for float numbers, it would be nice to add gorilla encoding.

**Reporter**: [Frank Dai](https://issues.apache.org/jira/secure/ViewProfile.jspa?name=soulmachine)

_{**Note**: *This issue was originally created as [PARQUET-2217](https://issues.apache.org/jira/browse/PARQUET-2217). Please see the [migration documentation](https://issues.apache.org/jira/browse/PARQUET-2502) for further details.*}

Start by reviewing Parquet's encoding specification and the existing BYTE_STREAM_SPLIT encoding support, then determine the format changes required for Gorilla encoding. Done means Gorilla encoding for floating-point numbers is specified and supported consistently across the format; the issue provides no file or test entry point.

Written by the indexing model from the issue text.