[AlsammanAlsamman]

Speed up allele comparisons and per-chromosome splitting

Results are unchanged; this is a pure performance change.

Where the time went

I profiled find_shared_dna and find_discordant_snps on two individuals with ~660k SNPs across 22 chromosomes and a HapMap2-sized genetic map. Most of the time was outside the genetic computation:

1. Allele-by-allele genotype comparison with the pandas .str accessor (.str[0], .str[1]): ~60% of find_shared_dna and ~88% of find_discordant_snps.
2. Repeated df["chrom"] == chrom scans: one full-table comparison per chromosome to split SNPs, plus _compute_shared_dna re-filtering its already single-chromosome DataFrame for every segment it saved.

Changes

• Alleles are decoded once per individual into an (n, 2) array of character codes (_get_alleles) and compared with NumPy (_no_shared_allele, _same_genotype, _is_discordant). A missing allele (single-allele calls, e.g. on X/MT) is code 0 and never matches anything, the same as the NaN produced by .str[1] before.
• SNPs are split by chromosome once with groupby("chrom", sort=False), which keeps the original chromosome order, and _compute_shared_dna uses its DataFrame directly.

Timings (best of 3, two individuals, ~660k SNPs)

	before	after
find_shared_dna	4.0 s	1.2 s
find_discordant_snps	1.1 s	0.24 s

Verification

• I compared outputs against main on 40 randomized datasets. The data covered unordered genotypes (AG/GA), single-allele calls, indel calls (ID/DI), a 3-character genotype, missing calls, X with a male individual, 2- and 3-individual comparisons, and pair and trio discordance. Every shared DNA segment table, discrepant SNP index and discordant SNP table was identical, including a run with low thresholds that produced ~19k segments per dataset.
• Added test_allele_comparisons for the helper edge cases. The existing tests pass.
• The 7 failures in tests/test_resources.py and test_sibling_shared_dna_proportions also happen on main against current snps main (Resources._download_file no longer exists). They aren't related to this PR.

This is independent of #129. If both are merged, the second one will need a small rebase around the match-mask code in find_shared_dna, and I'm happy to do it.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.13%. Comparing base (fa49eb4) to head (7fc3255).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #130      +/-   ##
==========================================
+ Coverage   97.04%   97.13%   +0.09%     
==========================================
  Files           5        5              
  Lines         575      594      +19     
  Branches       76       75       -1     
==========================================
+ Hits          558      577      +19     
  Misses          6        6              
  Partials       11       11              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[apriha]

Awesome work, thank you! Really nice speedup. 👍