fix(str): take the marker range from the majority value shape

[JamesKane]

Reported from the live report: DYS413 showed as multi-copy but with a single reported value.

DYS413  multi-copy  —  —   23   23   23   867  867  25      —  —  Not scored
                        min modal max              distinct

min = modal = max = 23 alongside 25 distinct values is self-contradictory.

Cause

DYS413 has 866 copy-vector calls (23-23, 21-23, 23-25, …) and exactly one bare 23 — a kit that collapsed the pair into a single count. The range picked whichever shape was non-null, with the scalar winning ties:

(Some(v), _) => v.to_string(),   // scalar wins, hides the vector

So one observation out of 867 replaced the entire copy-vector range, and the page presented a single sample's value as the corpus range. The mirror case was equally broken: a mostly-simple marker with one stray vector would have shown a one-sample vector range.

Fix

The range now comes from whichever shape the majority of observations use, and the other shape's statistics are nulled out, so a row can no longer carry both. DYS413 now reads:

	min	modal	max	notes
before	23	23	23	—
after	14-24	23-23	25-25	1 other shape

All three are real observations (14-24 ×1, 23-23 ×545, 25-25 ×1), and 23-23 is the genuine mode.

The minority observation is counted rather than silently dropped, and surfaced in the notes cell — same reasoning as null_alleles and complex_count, and it explains why distinct can exceed what the displayed range accounts for.

Fixed in the aggregation rather than the template, so /api/v1/reports/str-markers is correct too — it had the same bug.

Notes

• Migration 0072 adds mixed_shape_count. Existing rows keep the old values until the next du-jobs run-once str-marker-stats, which the deploy should run anyway.
• Regression test covers both directions (majority-vector with a stray scalar, and majority-scalar with a stray vector).

Verification

• cargo clippy --workspace --locked -- -D warnings clean
• du-db 115/115 against a fresh, empty database in parallel
• du-web 65/65
• Verified against the real corpus: DYS413's new endpoints confirmed as actual observations

🤖 Generated with Claude Code