compute_vecs_l2sq doesn't validate that vecs_l2sq.len() == num_points

[arrayka]

compute_vecs_l2sq doesn't validate that vecs_l2sq.len() == num_points. If a caller accidentally passes a smaller (or larger) output buffer, this will silently compute the wrong subset and still return Ok(()), which can lead to incorrect clustering/partitioning. Consider checking the output length and returning an error on mismatch.

Originally posted by @Copilot in #933 (comment)