selection(v2.2.3): no‑ML shortlist pipeline (brightness + spectral/family diversity)

[Mythmaker28]

Implements the lab-driven selection pipeline without ML, per your brief.

What’s included

• scripts/extend_table_qy_ec.py — adds QY, EC, brightness, photostability columns to the balanced table; brightness computed where both QY/EC present.
• scripts/select_shortlist_noml.py — selects 12 candidates by: brightness top‑30 → KMeans in (excitation, emission) → 1 per cluster → Calcium ≤3 → export CSV + spectral figure.
• reports/SHORTLIST_SELECTION_RATIONALE_v2_2_3.md — rationale and GO criteria.
• requirements.txt (pinned).

Does not touch master. Targets a dedicated branch.

Next

• Fill QY/EC from FPbase/papers (script currently leaves NA and computes brightness when available).
• Generate deliverables/lab_v2_2_3/ (shortlist_top12_final.csv, plate layouts, filters, protocol) once QY/EC are populated.

[Mythmaker28]

CI added for your no‑ML pipeline: .github/workflows/selection-no-ml.yml. It runs on this branch and on PR to master, builds the env, extends the table (QY/EC/brightness placeholders), executes the selection, and uploads artifacts (CSV + spectral figure).

Next step options:

• Provide/commit QY/EC values (or a mapping sheet) → extend_table_qy_ec.py will compute brightness for more rows.
• I can add a second score using measured contrast_normalized + family rarity weights (already implemented as fallback) to ensure non-empty shortlist even with few QY/EC.

As requested: no touches to master; all changes are isolated to selection/v2.2.3-noml.