Count data loading and genotyping initialization

[m-murphy]

Parent

#45

What to build

Extend the data-loading path so users can supply per-allele read counts instead of collapsing to presence/absence at load time.

Long-form loading accepts an aggregate argument ("binary" default | "count") and an optional reads column (default 1 per row when absent). Count mode sums reads per (sample, locus, allele) into non-negative integer barcodes. Delimited wide loading passes aggregate through after pivoting to long form. Long-form export / round-trip reflects counts when data were loaded in count mode.

The returned data object records aggregation mode (e.g. metadata field) so downstream code can distinguish count vs binary barcodes. Validation: clear error when aggregate = "binary" and duplicate (sample, locus, allele) rows exist; document that counts greater than 1 require count mode.

Update genotyping initialization so allele presence means count > 0 (not sum of read depths): observed COI hint, Jaccard similarity, and clustering-based initialization all use presence semantics consistent with count data. Uninformative loci (single allele) continue to be removed at load time.

Acceptance criteria

• load_long_form_data(..., aggregate = "binary" | "count") with optional reads column; default behavior unchanged (aggregate = "binary")
• load_delimited_data forwards aggregate to long-form loader
• Count aggregation produces correct summed barcodes; binary aggregation unchanged
• Returned data object includes aggregation mode metadata
• Duplicate-row error in binary mode with actionable message
• Long-form export preserves counts in count mode
• Genotyping observed-COI initialization uses alleles with count > 0 per locus (max over loci), not sum of depths
• Jaccard / clustering init treat allele as present when count > 0
• R testthat coverage in test-utils.R (or new test file) for aggregate, reads, validation, round-trip

Blocked by

None - can start immediately