ensure matching size schemas in filter_connected_components_by_size

[ptajvar]

Small fix to ensure that the schemas match when concatenating remaining component sizes and discarded component sizes in filter_connected_components_by_size.

---

Note

Low Risk
Localized graph filtering/statistics fix with a targeted test; no auth, data persistence, or API surface changes.

Overview
Fixes a Polars vertical concat failure when filter_connected_components_by_size merges in-component sizes with early-discarded component sizes that include an extra n_edges column.

discard_sizes from write_hive_partitioned_edgelist_without_out_of_size_bound_components carries component, n_umi, and n_edges, while sizes computed from the remaining edgelist only have component and n_umi. The filter now selects component and n_umi from discard_sizes before concatenation so pre-filtering component statistics stay correct without schema mismatch.

A regression test covers discard_sizes frames that include n_edges (e.g. multiplet recovery with max component size threshold).

^{Reviewed by Cursor Bugbot for commit 148a722. Bugbot is set up for automated code reviews on this repo. Configure here.}