feat: add donor-local inference and finalize 1.5.0

[Jaureguy760]

Summary

• Adds donor-local bulk allelic-imbalance orchestration for single|linear × global|per-donor inference.
• Keeps effect likelihoods and Benjamini-Hochberg correction donor-local. global shares only nuisance fits; per-donor fits nuisance parameters independently.
• Enforces unphased single-SNV testing while supporting separate phased and unphased peak-feature analysis.
• Aligns the Rust and Python APIs, type stubs, tests, documentation, changelog, citation metadata, and the 1.5.0 version surface.
• Rejects ambiguous multi-allelic count rows instead of coercing them into biallelic identities.

Validation

• make rust-test: 126 passed, 0 failed, 7 ignored.
• make test: 236 passed, 0 failed, 27 skipped.
• git diff --check origin/dev...HEAD: passed.
• Gitleaks: no leaks found across all eight commits.

Publication scope

• 19 source, test, documentation, and release-metadata files; no workflow, generated-analysis, or unrelated artifacts.
• No downstream SciLock analysis rerun.
• No merge, tag, package publication, or release action.