Implement lazy file reader capabilities

[d33bs]

This is a curiosity question, but could pycytominer work on backed data? Would this have speed up benefits? (probably worth discussing in the context of a narwhals upgrade in 2.0 as well)

Originally posted by @gwaybio in #573 (comment)

We could implement lazy file reader capabilities to enhance efficiency by avoiding full data reads unless absolutely necessary.

Examples / inspirations include:

• anndata.experimental.read_lazy (which enables lazy reads of anndata)
• pyarrow.parquet.ParquetFile (which abstracts Parquet data reads)
• pl.LazyFrame (for common interfacing with these data)
• narwhals.LazyFrame (for a common abstraction layer for multiple dataframe types)
• duckdb pl.LazyFrame support (another possible abstraction layer for lazyframes from a number of data sources)