feat: add predicate relationship metadata

[jonathanprozzi]

Status

Superseded by the classification-owned matrix approach in #7. This draft remains open only as historical review context for the predicate-level relationship experiment.

What This Experiment Tried

• Adds predicate-level relationship metadata keyed by subject classification slug.
• Adds expected object type hints and schema.org property names for a selected set of backfilled metadata predicates.
• Propagates relationship metadata through predicate records and covers the shape in registry tests.

Why It Should Not Merge As-Is

This shape puts classification-specific relationship metadata inside @0xintuition/predicates, which creates the coupling we want to avoid. It also mixes identifier namespaces in expectedObjectTypes and duplicates the subject/predicate relationship data that classifications now own through metadataPredicates.

The replacement direction is #7:

• @0xintuition/predicates defines reusable predicate vocabulary.
• @0xintuition/classifications owns the matrix keyed by (subjectClassification, predicate).
• Expected objects are typed as classification, schema, primitive, same-classification, or any.
• Schema mappings carry explicit match type: exact, semantic, broader, or narrower.

Stack

This branch is stacked on feat/predicate-metadata-backfill, but it should stay draft and out of the merge path unless we intentionally revive part of the experiment.

[jonathanprozzi]

Disposition note: this PR should stay draft / review-only and should not be merged as-is.

The relationship metadata shape explored here has been superseded by the classification-owned metadata predicate matrix in PR #7. That matrix moves subject/object relationship contracts out of predicate specs and keys them by (subjectClassification, predicate), which avoids predicate -> classification coupling and keeps object target namespaces explicit.

Useful review value from this PR: it helped identify the contract requirements. Merge path: #3 -> #4 -> #6 -> #7 -> #8, with this PR excluded unless we intentionally revive a narrowed piece later.