feat: Redesign mapper output to produce MappingRecord + Allele rows at all levels

[bencap]

Context

The updated MaveDB data model requires the mapper to output a MappingRecord + Allele rows at all applicable levels (genomic, coding, protein) for every variant — not just protein-level reverse translation. This is a broader interface change than the original scope of this issue.

Goal

Redesign the mapper output interface so that for each input variant, the mapper returns:

• A MappingRecord draft carrying provenance:
  • vrs_digest — VRS digest of the pre-mapped (assayed-level) variant, extracted as a top-level field for indexing
  • pre_mapped JSONB (raw input blob)
  • assay_level, mapping_api_version, QC fields (at_mismatched_locus, near_gap)
• A list of Allele draft objects — one per level — each containing:
  • level (enum: 'genomic' | 'coding' | 'protein')
  • transcript (NOT NULL — present for all levels)
  • The appropriate HGVS field(s): hgvs_g / hgvs_c / hgvs_p
  • vrs_digest (computed)
  • post_mapped JSONB (the raw mapper output blob for this allele at this level)
  • clingen_allele_id where already known (optional)

For protein-level score set targets, the existing reverse translation package is used to enumerate all coding variants encoding each protein change. Each produces a coding-level Allele draft. Non-protein-level targets (genomic, coding) produce no reverse-translated alleles but do produce alleles at all natively derivable levels.

Acceptance Criteria

• Mapper output schema includes a MappingRecord draft and a list of Allele drafts per input variant
• MappingRecord draft includes vrs_digest as a top-level field (not only buried in pre_mapped JSONB)
• All Allele drafts carry level, transcript, the appropriate HGVS field, vrs_digest, and post_mapped JSONB
• Protein-level mapping runs produce coding and genomic Allele drafts via reverse translation, in addition to the protein Allele draft
• Non-protein-level mappings produce alleles at all natively derivable levels
• Integration tests verify that mapping a protein variant returns the expected protein + coding allele drafts (one per codon variant per transcript)
• Integration tests verify non-protein-level mapping returns the expected allele set
• Downstream consumer (mavedb-api worker job, mavedb-api#740) updated to consume the new output schema