v0.3.3

Added

• Per-trait minor allele count (MAC) filter applied after sample subsetting, guarding against spurious p-values driven by singleton/very-rare variants when missing phenotypes or covariates reduce the cohort. Exposed as min_mac= on GWASPipeline.run_analysis() and PANICLE() (default 10 — twice the common PLINK --mac 5 convention, appropriate for inbred cohorts where effective sample size per allele is roughly half) and as --min-mac on the CLI. Set to 0 to disable.
• GenotypeMatrix.subset_markers() and GenotypeMap.subset_markers() helpers.
• Shared compute_mac_keep_indices() and pad_association_results() utilities in panicle.utils.stats.
• Per-trait Bonferroni denominator now uses the post-MAC marker count so the significance threshold reflects the number of tests actually performed.
• threads= parameter on load_genotype_vcf() and matching --threads CLI flag for tuning cyvcf2/htslib decompression workers (0 = all detected CPUs, default = min(4, cpu_count)).

Changed

• Default behavior: min_mac=10 is now applied by default in the high-level GWAS APIs. Pass min_mac=0 to restore pre-0.3.3 behavior.
• VCF first-load ingestion now writes the dynamic int8 matrix in marker-major order so each appended marker is a contiguous write, then transposes once on finalize. Speeds up VCF first-load without affecting the cached output layout.
• load_genotype_vcf(backend='auto') now prefers cyvcf2 when installed (previously defaulted to the builtin text parser for VCF/VCF.GZ).