Make VCF header-field matching data-driven (field as selector), aligning VCF with the SAM header_field design

[NoopDog]

Context

Surfaced during self-review of #221 (PR #234). match_vcf_header_pattern bakes the matched field into code, one canonical field per header type:

• ##reference/##source -> the value
• ##INFO/##FORMAT/##FILTER -> the ID field
• ##contig -> the assembly field (added in VCF ##contig assembly= rules can never match (scanner splits fields, patterns expect key=value) #221)
• other -> the raw line

So a rule can only ever match the one hard-coded field per type. Matching a different ##contig subfield (e.g. species, md5) would require a code change.

The asymmetry

The SAM side is already data-driven — the field is a parameter and rules declare it:

match_sam_header_pattern(header, section, field, pattern)   # SAM: field is a PARAMETER (rules carry header_field: PL/SN/AS/...)
match_vcf_header_pattern(header, header_type, pattern)      # VCF: field baked in per type

Proposal

Make the VCF matched field declarative, mirroring SAM. Either a vcf_header_field key in the rule when (preferred, matches the SAM header_field convention) or a type.field selector form (e.g. ##contig.assembly). The engine's _match_vcf_header would pass the field through, and match_vcf_header_pattern would look it up instead of hard-coding it.

Interaction with #154 (READ BEFORE STARTING)

#154 (rules with fact-based antecedents: extractors + derivers + rules) is the umbrella this sits under. If #154 introduces extractors that emit named facts and rules that match facts, the entire regex-against-a-named-header-field mechanism — both today's and this proposal's — is subsumed: rules would match a reference_assembly fact, not a regex against ##contig.assembly.

Therefore:

• This is a safe, small consistency win if done before Design: rules with fact-based antecedents — collapse the six per-format classifiers into extractors + derivers + rules #154.
• It is interim work Design: rules with fact-based antecedents — collapse the six per-format classifiers into extractors + derivers + rules #154 would replace if the larger redesign lands first.
• Sequence deliberately: if Design: rules with fact-based antecedents — collapse the six per-format classifiers into extractors + derivers + rules #154 is being actively designed, prefer folding this into it rather than shipping an interim field-selector.

Related deriver work: #236 (accession->assembly) is itself a #154-style deriver.

Definition of done

• Decide the declarative form (vcf_header_field key vs type.field selector); prefer parity with SAM's header_field.
• match_vcf_header_pattern looks up the declared field; engine _match_vcf_header passes it through.
• Migrate the existing VCF header rules (contig/INFO/FORMAT/FILTER) to declare their field.
• Rule schema/validation accepts and (where possible) validates the field.
• Tests updated; make test-all, make lint, make format-check, make type pass.

Related

• VCF ##contig assembly= rules can never match (scanner splits fields, patterns expect key=value) #221 / PR Fix VCF ##contig assembly rules to read the parsed assembly subfield (#221) #234 (baked-in ##contig assembly field; surfaced this)
• Design: rules with fact-based antecedents — collapse the six per-format classifiers into extractors + derivers + rules #154 (extractors + derivers + rules — the umbrella; may subsume this)
• Resolve reference assembly (and patch) from GenBank accession via a deriver, not regex ranges #236 (accession->assembly deriver)