-------------
Improvements
~~~~~~~~~~~~
- Improve performance of report extraction and features calculation by writing partial DataFrames to the shared memory (or temp directory, if shared memory is not available).
Both the used memory and the execution time should be lower than before, when processing large DataFrames.
- Use zstd compression instead of snappy when writing parquet files.
- When ``repo`` is pickled, extract the DataFrames only if they aren't already stored in the cache.
- Remove fastparquet extra dependency.