Disjunction (OR) in rule bodies

[oflatt-claude]

Adds disjunction to rule queries:

(rule ((OR ((edge x y))
           ((edge y x))))          ; symmetric closure
      ((connected x y)))

Each branch of an OR is a parenthesized list of facts (a conjunctive subquery); the rule matches when any branch matches. Only variables bound in every branch may be used in the actions.

Status: draft — establishes the contract, backend is being rewritten

This PR currently lands the syntax, semantics, behavioral tests, and design doc. The current backend is a parse-time rule-splitting prototype (it distributes the body into the cartesian product of branch choices, one plain rule per combination).

Per review, rule-splitting is not the target — it has combinatorial blowup and redundant (idempotent) firing. The implementation will move to native execution:

• Strategy B — materialized union subquery: compile OR to a derived relation R_or(V) = ⋃ᵢ π_V(branchᵢ) over the common variables V, materialized via the existing hypertree-decomposition path in core-relations/src/free_join/plan.rs, then joined as one atom. No blowup, auto-dedup.
• Strategy C — native union node in the free-join executor: streaming, no materialization; more invasive.

See docs/disjunction-design.md for the full comparison, the seminaive/incremental-maintenance considerations, and code references.

What's here now

• src/ast/parse.rs — parse (OR ...) and expand a body; also fixes a latent todo!() when a fail sub-command expands to several commands.
• tests/disjunction.rs — 6 behavioral tests (symmetric closure, cartesian product, nested OR, single-branch, branch-local-var rejection, empty-OR rejection). These are implementation-independent and will guard the B/C rewrite.
• docs/disjunction-design.md — syntax, semantics, and efficient-execution design.
• CHANGELOG.md — feature entry.

Tests

• tests/disjunction.rs: 6/6 pass
• full .egg harness: 747/747 pass (desugar, 32-thread, proof treatments)
• make nits: clean

🤖 Generated with Claude Code

[oflatt-claude]

Superseded by #7 (the chosen implementation: correlated + seminaive fused OR). Kept the exploration history here; closing to leave the two deliverables (#7 + the encoding change) as the live PRs.