add multi-target regression support

[azmyrajab]

This PR adds support for multi-target regressions, where:

1. user is expected to pass a polars expression returning a polars struct for "targets". This struct contains the names and values of every target they wish to regress.
2. *features, ols_kwargs, and other parameters remain the same. The Rust implementation will convert the target series struct to a 2D ndarray and compute validity masks across all unnested target and feature series
3. LAPACK DGELSD or pure rust ridge SVD implementation is used to factorize "X" and solving against multiple 2D Y once

Notebook, demo benchmark, and tests are updated with examples.
For large number of targets, it is more efficient to use this multi-target functionality than to run multiple indepedent linear regressions.