Use polars for collapsing archive OTU tables

[wwood]

Motivation

• The write_collapsed_paired_with_unpaired_otu_table implementation used pandas and produced failures in the collapse-to-sample tests; switching to polars enables safer, typed ingestion and deterministic grouping/aggregation for archive OTU tables.

Description

• Replaced pandas usage in write_collapsed_paired_with_unpaired_otu_table with polars and removed the pandas import.
• Added archive_schema to map archive fields to typed Polars schema and read rows with orient="row" to avoid orientation warnings.
• Concatenated archives with pl.concat, removed sample suffixes using pl string ops, checked unique constraints via Polars, and sorted by sequence/gene to preserve deterministic order.
• Collapsed groups using partition_by and manual aggregation to preserve list-valued fields (read_names, nucleotides_aligned, read_unaligned_sequences, equal_best_hit_taxonomies) and produce ar.data as the expected list-of-lists structure.

Testing

• Ran pixi run -e dev pytest test/test_summariser.py which passed: 28 passed, 1 skipped.
• Ran the full test suite with pixi run -e dev pytest test, which completed but had one unrelated failure in test/test_pipe.py::Tests::test_sample_name_strange_characters caused by an external diamond process exit (exit status -7), not by the summariser changes.
• Iterative test-fix cycles were performed while developing the change until the summariser tests were green.

---

Codex Task