Zstd compression can yield superior compression as compared to Snappy on some types of input data.

• Add a new ParquetWriter factory method to enable user selection other than the default.
• Add method to read buffered+written data size of the underlying Writer to support occasional external flushing
• (Somewhat crudely) Update the single unit/integration test to ensure both Snappy and Zstd function and that the codec's name appears in the first metadata block