refactor(legacy_expr): step γ1 — Aggregate bridge module + output_schema helper

[zzylol]

Summary

First sub-PR of Step γ (variant-by-variant consumer migration). Adopts the (c) bridge strategy: keep legacy_expr::QueryExpr::Aggregate as the L2 emit shape; add a one-way canonical-builder helper consumers call on demand. No construction site rewritten; no bridge enum variant added to legacy QueryExpr.

This avoids the "child must be canonical" coupling that would otherwise force γ1 to migrate every other legacy variant (SketchAgg / WindowedAgg / TopK / …) in the same PR. The bridge module + output_schema_for_aggregate are the foundation γ2-γ4 will consume.

What landed

New controller/src/intent_algebra/aggregate_bridge.rs (328 lines, 8 tests)

pub fn bridge_aggregate_to_canonical(
    keys: &[ColumnRef], aggs: &[AggItem],
    having: &Option<legacy::Predicate>, schema: &Schema,
) -> Result<BridgedAggregate, BridgeError>;

pub struct BridgedAggregate {
    pub by: Vec<ColumnId>,
    pub aggs: Vec<AggIntent>,
    pub having: Option<HavingPredicate>,
}

pub enum BridgeError {
    UnresolvedKey(ResolveError),
    HavingDeferred(QueryExprError),
}

The having translation routes through from_legacy_scalar (Batch 2). E-deferred ScalarExpr variants surface as BridgeError::HavingDeferred(...) — test bridge_having_deferred_e_variant_surfaces_error is the contract.

column_resolution.rs extension (+211, +6 tests)

pub fn output_schema_for_aggregate(
    input: &Schema, by: &[ColumnId], aggs: &[AggIntent],
) -> Schema;
pub fn resolve_named_keys(keys: &[ColumnRef], schema: &Schema)
    -> Result<Vec<ColumnId>, ResolveError>;

Mirrors canonical query_expr::QueryExpr::output_schema_in's Aggregate arm. This was Step β TODO #1. γ2-γ4 consumers descending into legacy Aggregate.input should pass the output schema as the inner subtree's parent_schema.

Demo wire in physical/allocator.rs::alloc_node

The legacy Aggregate arm calls the bridge to enrich NodeAnnotation.rationale with intent kinds + group-by column count. Emit shape stays legacy; behavior unchanged. Proves the bridge is reachable.

Construction-site migration: 0 (intentional)

Per strategy (c). γ7 flips construction sites once γ2-γ4 retire legacy SketchAgg / WindowedAgg / TopK — so no legacy input subtree remains.

Build + test

• cargo build --release -p controller clean
• cargo build --release -p query_engine_rust clean
• cargo test -p controller --lib — 688 passed (was 674; +14 new bridge + column_resolution tests)
• cargo test -p controller --bin controller — 27 passed

Caveats

• Canonical QueryExpr::Aggregate.having is Option<HavingPredicate> (HavingPredicate(pub String)), NOT typed Option<Predicate> as spec suggested. Bridge renders converted Predicate via format!("{pred:?}"). Roundtripping is a follow-up.
• BridgeError can't derive PartialEq (QueryExprError doesn't); tests use matches!.
• optimizer::engine::HydraConversion (Step β TODO test: e2e test for modified-OTLP CountMin sketch hot path (PR D) #7) NOT migrated — deferred.

Diff: 5 files, +618 / -4

🤖 Generated with Claude Code