Introduce an interface for controlling the encoding per column

There are use-cases in which it is advantageous to have control over the type of encoding used for a given column. To achieve this, the ParquetProperties class must be refactored to facilitate its subclassing. The constructors of ParquetWriter and (probably) InternalParquetRecordWriter have to be modified as well. All the internal parquet classes have to refrain from directly instantiating ParquetProperties and should accept the instance from outside.

Currently me and my team are using a good amount of reflection and unenforceable assumptions in order to gain control of this behavior. I'd like to implement the interface change. Are you interested in this feature?

**Reporter**: [Boris Peltekov](https://issues.apache.org/jira/secure/ViewProfile.jspa?name=peltekster) / @peltekster

_{**Note**: *This issue was originally created as [PARQUET-226](https://issues.apache.org/jira/browse/PARQUET-226). Please see the [migration documentation](https://issues.apache.org/jira/browse/PARQUET-2502) for further details.*}

Start by reading ParquetProperties and the constructors of ParquetWriter and InternalParquetRecordWriter. Trace internal Parquet classes that instantiate ParquetProperties, then define the interface and dependency flow needed to control encoding per column; done means callers can supply the encoding behavior without reflection or direct internal instantiation.

Written by the indexing model from the issue text.