Evaluate polars as a replacement of pyspark

[kpto]

Following #18, Polars is preferred as it's syntax resembles Pandas and perhaps easier for a wider audience who may not have much understanding in SQL preferred by duckdb. A minimal working sample will soon be implemented and findings will be recorded here.

Initial findings regarding Polars:

1. The support of streaming was added recently and is not mature https://docs.pola.rs/api/python/stable/reference/lazyframe/api/polars.LazyFrame.collect.html#polars.LazyFrame.collect
2. Unlike duckdb which uses disk space to handle memory heavy operators, Polars seems to have no corresponding measure.
3. The above requires developers to understand the memory usage of operators well to avoid potential memory issues. This is a good reference: https://www.rhosignal.com/posts/streaming-operations-in-polars/

[kpto]

Unfortunately the streaming feature seems to be very new and not suitable for production use. The doc is still under construction:
https://docs.pola.rs/user-guide/lazy/streaming/

Also it's unclear on how to iterate rows in Python runtime, the argument streaming=True of collect turns on streaming but it still returns a DataFrame which does not seem to be a generator.
https://docs.pola.rs/api/python/stable/reference/lazyframe/api/polars.LazyFrame.collect.html#polars.LazyFrame.collect

Here is also a relevant issue showing that a LazyFrame cannot/inefficient to be iterated in Python runtime:
pola-rs/polars#10683

Polars remains in our watchlist but not suitable to use for now. Closing the issue.