infer-roothogs silently drops proteins #114

Problem

Some proteins of the input proteomes never reach a rootHOG FASTA (and so no output file), without any warning. Unlike #113 this has not been quantified on fungi-30 yet.

• merge_rhogs2 overflow: when a merge cluster exceeds --big-rhog-size, newhost is not reset between clusters and the overflow host loses its own proteins (_utils_roothog.py:667-713). Also, if all members of a cluster are gone, the previous cluster's host_hog is reused.
• Proteins without any hogmap row are dropped (parse_hogmap_omamer, _utils_roothog.py:264). They are neither mapped nor in unmapped, so they don't reach linclust either.
• Ids longer than 230 characters are truncated (_utils_roothog.py:256) and then no longer match the hogmap qseqid (:289). check-input rejects ids longer than 60 characters, so this only happens when check-input is bypassed; the two limits should agree anyway.
• Minor isoforms reach linclust: isoform selection filters the hogmaps but not the unmapped proteins, so non-selected isoforms are clustered (infer_roothogs.py:50-60, :93).

Tests (strict xfail on PR #112)

• tests/test_roothog_grouping.py::TestMergeRhogs2::test_overflowing_clusters_keep_all_proteins
• tests/test_roothog_stage.py::test_synthetic_protein_missing_from_hogmap_not_lost
• tests/test_roothog_io.py::TestParseHogmapOmamer::test_long_ids_are_kept
• tests/test_roothog_stage.py::test_synthetic_minor_isoforms_not_clustered

Proposed fix

Fix the four cases, and add an accounting step at the end of fastoma-infer-roothogs: every protein of the proteomes (after length filtering and isoform selection) is either in a rootHOG FASTA, in singleton_unmapped.fa, or reported as a deliberately dropped singleton, and the counts are logged.