Fix query-engine bugs: :in predicate bindings, intra-tx lookup refs, EDN symbol chars

[tiensonqin]

Summary

Fixes several engine bugs hit by the Logseq db-worker port, root-caused in the engine and verified against upstream juxt/datascript semantics. New test suite test_query_engine_fixes covers each fix (upstream tests ported where they exist; aliases added to upstream_test_aliases.tsv).

Bug: [(= ...)]/[(not= ...)] predicates over :in-bound vars drop all rows — CONFIRMED + FIXED.
bound_relation_clause in impl/query_where.ml only substituted bindings into patterns and 2-arg ComparisonPredicate, leaving EqualityPredicate/ComparisonPredicateN untouched. A :in-bound var stayed a QVar, so filter_relation_equality couldn't resolve it and every row was dropped. Now all relation-evaluable clause kinds get bound: equality predicates, N-arg comparison predicates, arithmetic/name/namespace/keyword value clauses, SourceClause, and Not/SourceNot/NotJoin/SourceNotJoin (the function is now rec). [(= ?v ?target)] and [(not= ?d ?x)] over :in scalars now return the right rows.

Bug: intra-transaction lookup refs in entity maps fail — CONFIRMED + FIXED.
apply_entity_map called resolve_entity_attrs which strictly resolved all attr values against the datoms accumulated before the entity map ran. An attr value like [:block/uuid #uuid "..."] referencing an entity created by an earlier attr of the same map (or an earlier tx op) raised "Nothing found for entity id". Upstream transact resolves each [:db/add] op sequentially against the accumulating db-after. Now the entity-map fold resolves each tx_value strictly inside apply_attr at its add step, so earlier attrs of the same map are visible; the upsert probe (entity_unique_identity, validate_explicit_upsert_target, tuple_identity_lookup_writes) uses a non-strict resolution matching upstream resolve-upserts (unresolvable refs keep raw form and never match). Verified: earlier-op refs, same-entity self-refs, Add-op refs to earlier-tx entities, tempid/uuid upsert merging all work; forward refs and unresolvable refs still correctly raise.

Bug: EDN reader can't tokenize ' inside symbols/keywords — CONFIRMED + FIXED.
' was in read_edn's is_delimiter set, splitting foo' into foo + quote. Upstream EDN treats ' as a non-terminating macro char: special only at token start, legal mid-symbol. Removed from the delimiter set; leading-' quoting still works via parse_form. {:user.property/foo*+!_'?<>=- nil} and [sym' foo*+!_'?<>=- 'quoted ?var] now parse like cljs edn/read-string.

Bug 1 (recursive rules wrong direction): not reproducible — ported upstream test-rules cases (recursive walk, "Rule with branches" positional binding with renamed head vars, swapped-arg symmetric closure) all pass on main. Kept as regression tests.

Bug 3 ([?x ...] collection :in ignored): not reproducible — collection inputs iterate correctly via the sequential clause path; ported upstream test-bindings collection case kept as regression test.

Bug 4 (avet not backfilled on kvs sqlite restore): not reproducible — restore paths rebuild avet (existing sqlite test asserts datoms restored Avet; memory/file storage verified empirically: attr+value lookups, lookup-ref resolution, and avet queries all work post-restore).

Full dune runtest is green with upstream parity checks enabled.

Link to Devin session: https://app.devin.ai/sessions/c82925b6976a4a6da2688c410e65011a
Open in Devin Desktop: https://app.devin.ai/desktop/session/c82925b6976a4a6da2688c410e65011a?variant=devin
Requested by: @tiensonqin

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.