fix(query): accept Bool, Date, and DateTime in min and max

[azimafroozeh]

What & why

Closes #623. min($p.birthdate) over a Date column was refused at typecheck (T8: min requires numeric or string type, got Date?), and DateTime and Bool columns the same way; only numeric and String columns passed, and the executor had arms for those alone. Found converting Kuzu's aggregate tests to .gqt cases.

• The T8 rule for min/max keys on one predicate naming the accepted scalars (numeric, String, Bool, Date, DateTime) and its message lists them; lists, vectors, and Blob values stay refused.
• The executor gains Boolean, Date32, and Date64 arms through the existing typed min/max path (false before true, dates chronological); the result keeps the column's own type, as before.
• The zero-row shortcut that typed every non-count aggregate as a Float64 null is deleted: an all-aggregate query over zero matched rows now flows through grouping as one empty group, so count is 0 and min/max return a null of the column's own type; sum and avg still build Float64.
• Case issue_623_min_max_over_date_and_bool.gqt: one accept step over all three column types, a zero-row step and a null-only-row step whose --- expect shape blocks pin the result types through the schema check from test(gq): add a mandatory --- expect shape section and a result-schema check #635, and four refusal steps (min over a list, max over a Vector, sum over a Date, min over a Blob).

Backing issue / RFC

• Fixes an accepted issue: Closes bug: min and max refuse Date and Bool columns #623 (the diff carries issue_623_min_max_over_date_and_bool.gqt)

Checklist

• Change is focused (one typecheck predicate, three executor arms, and the zero-row builder that contradicted the declared type)
• Tests added/updated for behavior changes (the .gqt case, 7 steps: accept, zero rows, null-only row, four refusals; result types via --- expect shape; no Rust test, nothing the format cannot express)
• Public docs updated if user-facing surface changed (docs/user/queries/index.md: accepted types, ordering, zero-row shape; docs/releases/v0.11.0.md: Highlights bullet)
• Reviewed against docs/dev/invariants.md — no Hard Invariant weakened, no deny-list item hit (typecheck and projection only; no storage, commit, or policy path touched)

Local verification

• cargo test -p omnigraph-gqt — 116 unit + 12 cases green on 50f3a6bc (the new case: 7 steps)
• cargo test -p omnigraph-compiler — 332 green
• cargo test -p omnigraph-engine --test aggregation — 8 green
• cargo clippy --workspace --all-targets — green
• cargo fmt --all --check — clean
• python3 scripts/check-docs.py — 126 files OK
• cargo test --workspace — run on the previous base 9570c7a7: lib 403 passed / 1 failed (external_blob_file_policy_rejects_special_files, a known failure under the dev-machine sandbox), 45 integration binaries green; not rerun on 50f3a6bc

Notes for reviewers

• Behavior change beyond the accept set: over zero matched rows an all-aggregate query's min/max columns are now typed as the column, not Float64. A JSON row omits a null cell's key, so the difference shows only in the Arrow IPC result schema from feat: unify graph JSON output on one Arrow writer and offer Arrow IPC as a query result format #627.
• The deleted zero-row builder was a second producer of that row shape; the grouping path already builds it, and a no-match query that also projects a group value still returns zero rows (stated in the user doc).
• Bool in min/max follows Kuzu, Neo4j, DuckDB, and Memgraph; PostgreSQL refuses it and offers bool_and/bool_or instead.
• Pre-existing and out of scope: the T8 rule consults the accept set only for scalar arguments, so min($p) over a node variable still typechecks and orders the id string.
• Drive-by in the same user-doc paragraph: the stray '''''' left by fix(compiler): refuse two projections that execute to one result column name #621 is removed.

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.