v1.22.3

Expression datasets now expose optional assignment sources through evidence_sources() and with_evidence(source_id, feature_ids=...). The dog integration consumes ORFdata's pinned, released peptide evidence, preserving all candidate proteins, reported memberships, native confidence/exclusion fields and exact source identities in portable exports. Values, missingness, contracts and donor counts are unchanged.

The optional dataset is discoverable/acquirable through the existing oncoref data commands. Enrichment defaults to offline use with a zero transfer budget; ordinary reads remain summary-only. The full optional release transfers 6.9 GB and occupies 24.7 GB installed. Bounded validation uses the producer's audited native/synthetic fixture, avoiding a large local download.

PR: #646
Closes #585 after PR #645.
PyPI: https://pypi.org/project/oncoref/1.22.3/

Validation: all eight CI checks passed, including Python 3.10–3.14, minimal installs and Windows. 134 focused local tests, lint and strict docs passed; 13 evidence checks passed from the extracted source archive. A normal wheel with public ORFdata 0.42.2 produced a 130,566-byte selected-evidence export that replayed without ORFdata or PyEnsembl. The exact tested distributions were published after comparing package contents and merged/tested trees. Attached receipts retain source/artifact hashes and acceptance scope.