fix(mimic-iv): update Charlson ICD-10-CM codes for C4A, C7A, and C7B

[acnimma]

Summary

• Exclude C4A, include C7A in malignant_cancer, C7B in metastatic_solid_tumor
• Regenerate postgres and duckdb copies from BigQuery source

Test plan

• pytest tests/test_transpile.py (161 tests)
• Transpiled postgres/duckdb charlson.sql
• CI concept build on MIMIC-IV demo

Fixes #2017

[Chessing234]

c4a/c7a/c7b handling is the right call for quan 2005. worth a short comment in the sql so the next person doesn't "fix" it back in.

[acnimma]

Quan notes in charlson.sql thanks

[Chessing234]

i checked the code ranges rather than taking them on trust, and the two additions do hold up: 'C7A' sorts above 'C76' and below 'C81', so it genuinely falls outside every existing range, and 'C7B' isn't in the IN ('C77','C78','C79','C80') list. the C4A exclusion is also doing real work — 'C4A' sorts between 'C45' and 'C58', so without it Merkel cell would have been picked up by that range. so the logic is sound and the operator precedence works out the way you intended.

two things before this lands.

the postgres copy isn't transpiler output. running the repo's transpiler (sqlglot 30.11.0 from requirements-lock) against your bigquery source and diffing mimic-iv/concepts_postgres/comorbidity/charlson.sql on this branch, sqlglot emits the reference header as a single line and the branch has it wrapped across 16. the duckdb copy matches exactly, and the same check on main is identical, so this branch introduces it. generated-up-to-date in transpile.yml compares that folder against transpiler output.

the one i'd actually push back on: excluding C4A while C43 stays in is inconsistent. your comment says C4A is excluded "as skin malignancy per Quan", but Quan's ICD-10 list explicitly includes C43 (malignant melanoma of skin) — main already has it in the IN ('C43','C88') branch. the Charlson convention is to exclude non-melanoma skin cancer (C44) and count melanoma. Merkel cell is an aggressive cutaneous malignancy, much closer to melanoma than to C44, so counting C43 but not C4A is hard to defend on the rationale given. i'd either drop the C4A exclusion or explain why it's grouped with C44 rather than C43.

and the evidence question: Quan 2005 predates all three codes, so this is a deliberate extension of the published algorithm rather than a correction to it — worth saying so in the body. could you also show how many hadm_ids actually change? C4A/C7A/C7B entered ICD-10-CM well into MIMIC-IV's coverage window, and if they appear in diagnoses_icd a handful of times or not at all, that's worth knowing before changing a widely-used comorbidity concept.