[Paururo]

Follow-up to #18, which told gaps never searched apart from searched empty ones in the states file and the report. Two places downstream still treated them as audited absences.

Evolutionary audit. Gains and losses are made of absences, and evolutionary_events.tsv built them whether or not Recore had searched those gaps; under the default --min-prevalence, every gain on a cluster in fewer than one genome in ten read as an audited call.

• New columns absent_not_searched and not_searched_reason (low_prevalence, no_cluster_sequence, no_fasta).
• For calls that rest on absences (acquisition, reference loss, deletion loss, homoplastic, unresolved): when none of the absences behind the call were searched (for a reference loss, when the reference genome's was not), support drops one level, so score and priority follow, and mechanism_detail gets absences_not_searched. When only some were, it gets absences_not_searched=N/M and no downgrade. Either way recommended_wording says so.
• Calls made from rescue evidence (pseudogenes, fragments, paralogs, reference events) are untouched.

Run comparison. benchmarks/compare_runs.py counted any state other than ABSENT as a call, so comparing against a run from before #18 listed every NOT_SEARCHED_* cell as a new correction. They are absences now, and a cell searched in one run and not the other is search_coverage_changed, counted apart from changed_cells.

Checked by running it

• fmt, clippy -D warnings, 187 library tests, 3 integration tests with minimap2 required, mkdocs build --strict.
• The new unit and end-to-end tests fail when the downgrade, the count or the qualification is removed.
• Tutorial: Rescued 11 and Matrix-corrected 8 unchanged; only Rv3810 (below the floor) gains the tag and wording, keeping its priority.
• Tutorial without the L5_AS00501 FASTA: Rv0008c, whose only absence is L5, goes from moderate to low support and from highlight to candidate (score 0.884 to 0.664); Rv1180, Rv1181 and Rv2930 are tagged 1/9 or 1/8 and keep their calls.
• compare_runs.py, tutorial output against a copy with ABSENT in place of NOT_SEARCHED_*: 11 new corrections before the fix, 0 after, with 11 search_coverage_changed; a run against itself is all same.