[dimitri-yatsenko]

Docs for datajoint/datajoint-python#1640 and its fix, datajoint/datajoint-python#1642.

Lineage follows foreign keys: a table's native primary-key attributes start it, and foreign keys pass it on. An attribute that no foreign key brings in has none: a native secondary attribute, an attribute computed in a projection, or a key that dj.U promotes. A join on two such namesakes is refused as a join on unrelated attributes, not as missing lineage.

• Explanation, semantic-matching.md:
  • The lineage rules say where lineage comes from, and add keys promoted by dj.U.
  • The matching rule "either lineage is null → cannot verify homology" now reads as unrelated namesakes.
  • A new section, "Joining on unrelated namesakes", shows the error with two tables that each declare input_hash, and how to rename or exclude it.
  • The page no longer sends users to rebuild_lineage() for this error.
• Spec, reference/specs/semantic-matching.md:
  • The "Raises" lines and the lineage-population notes now cover attributes with no lineage.
  • Both new messages are documented under Error Messages, marked "Changed in 2.4".
• Union, corrected in both pages. The pages said union requires matching lineage. Union never compares lineage: it requires the same primary key and no secondary attribute in common. The docs now say what the code does.

Merge with or after datajoint/datajoint-python#1642.