Optimization: Push down projection and filter predicates down to scan from source

[ray6080]

Description

For simple cases, like LOAD FROM 'a.parquet' WHERE a.age < 10 RETURN a.name, we should be able to prune unrelated columns and push the filter down when scanning from the parquet file.

This should be generalized to other sources, such as Arrow tables, Polars & Pandas DataFrames, too.