h2oai/datatable ? reverse-engineered prompt

Build me a Python package for working with large tabular data, kind of like a faster alternative to pandas for data frames.

I want it to handle columns efficiently, support strings, dates, categories, null values, and work well on big files without loading everything into memory at once. It should be able to read CSVs quickly, sort, group, join, filter, and do common data cleaning tasks with as little copying as possible. Make it use multiple cores when that helps, and let users switch to or from pandas, numpy, pyarrow, or plain Python data easily.

Please include a simple, friendly API, good tests, and basic docs so someone can install it with pip and try it right away. If you need to check current best practices or external docs while building it, go ahead and look them up online.