Compute a canonical disease display label in the ETL

[zuhdil]

Summary

Compute a single canonical disease display label in the ETL and carry it
into Gold as a new dim_disease.disease_label column. This is the data-layer
half of the client's final disease-name rule; the dashboard reads this column
(separate PR in igh-dashboard).

The rule

With P = disease_filter and S = secondary_disease_name (the
"No secondary disease" sentinel and blanks are already normalised to NULL
upstream):

• Malaria → P + " – " + S (e.g. Malaria – P. falciparum) — its strains
  never stand alone, so the primary is prefixed with a spaced en dash.
• secondary present → S (COVID-19, Lassa fever, Gonorrhea).
• no secondary → P (Tuberculosis, Buruli ulcer).
• both missing → NULL (the cross-cutting "R&D for all" rows).

STIs need no special branch: they arrive parent-collapsed with the specific
infection in secondary_disease_name, so the default branch prints it.

Why compute it here

Computing the label once in diseases.py, beside the existing secondary-sentinel
and STI normalisation (and unit-tested alongside them), keeps the rule in a single
tested place. The downstream GraphQL backend and React tables/slide-ins become
pure pass-throughs with no duplicated logic.

Changes

• transformations/diseases.py — compute disease_label after the existing
  disease-filter normalisation.
• silver_to_gold/config/schema_map.py — pass disease_label through to
  dim_disease.
• tests/unit/test_diseases.py — cases for the default, no-secondary, Malaria,
  STI, and both-null branches.
• tests/e2e/test_silver_to_gold_e2e.py — add disease_label to the expected
  dim_disease column set.

Verification

• uv run pytest tests/unit/test_diseases.py — 22 pass.
• Regenerated Gold and spot-checked: Malaria rows render Malaria – <strain>,
  secondary-bearing rows render the secondary, no-secondary rows render the
  primary group, 239 cross-cutting rows are NULL.