refactor(legacy_expr): batch 2 — lift 10 A-variants into canonical L3 + typed Predicate

[zzylol]

Summary

Additive lift of all 10 A-classified legacy variants into the canonical intent_algebra::query_expr::QueryExpr, plus the minimal typed Predicate IR for Filter.pred. Zero consumer-site redirects — those happen in subsequent C-reshape batches.

What landed

Canonical QueryExpr grew 5 → 15 variants

Pre-existing (PR #130)	Added in this PR
Scan, Window, Aggregate, LetBinding, Ref	Filter, Project, Partition, Distinct, Merge, Join, SetOp, Sort, Limit, BinaryOp

Naming: single-input variants use child: (canonical convention). Multi-input variants follow design.md §6 shapes verbatim.

Supporting types

ColumnRef, PartitionKeys, BinaryOpKind, JoinKind, SetOpKind, SortKey, VectorMatch{,Kind}, VectorGrouping, GroupSide, LiteralValue, ProjectItem — all lifted into canonical + re-exported from intent_algebra::mod.rs.

New typed Predicate

pub enum Predicate {
    Column(ColumnRef),
    Literal(LiteralValue),
    BinaryOp { op: BinaryOpKind, lhs: Box<Predicate>, rhs: Box<Predicate> },
    IsNull { expr: Box<Predicate>, negated: bool },
}

Plus Predicate::from_legacy_scalar(&legacy::ScalarExpr) -> Result<Predicate, QueryExprError>. Translates the 4 supported variants; returns QueryExprError::UnsupportedLegacyScalar(name) for the 4 E-deferred (FunctionCall / ScalarSubquery / InList / Between) per user decision.

Handles LiteralValue::Duration → Int(nanos) fold (canonical LiteralValue deliberately narrower).

Why zero consumer redirects in this PR

Agent's honest finding: every legacy-A-variant consumer (query_parser/, physical/{stage_split,planner,allocator}.rs, optimizer/engine.rs, legacy_lower.rs) simultaneously matches A-variants AND C-variants (Source, Aggregate, Window, SketchAgg, WindowedAgg, TopK, etc.) AND constructs ScalarExpr predicates with E-deferred variants. Migrating them mid-pipeline would require lifting C-variants and E-variants in the same PR — out of scope per the batch plan.

Consumers migrate one C-batch at a time:

Batch	What
3a (next)	AggFunc → AggIntent (medium)
3b	TopK split into AggIntent::TopK vs Sort+Limit (medium)
3c	WindowedAgg un-fusion — 20+ consumers; HIGH risk
3d	SketchAgg → Aggregate@L3 + L4 binding reads canonical; HIGH risk
3e	PromQL parser-side histogram_quantile → Quantile substitution + drop legacy HistogramQuantile/PromQLSubquery (low/mechanical)
13	Retire legacy_expr.rs + legacy_lower.rs

Catch-all arms in 6 canonical-side consumers

Files: optimizer/cost/mod.rs, physical/colored_dag/{allocator,emitter}.rs, sketch_algebra/lower.rs, warm_tier_analysis.rs. These previously matched canonical QueryExpr exhaustively over the 5 pre-existing variants; now they handle the 10 new ones conservatively (Edge stage / Logical wrap / child-walk / zero-cost) with TODO markers.

Serde caveat

Initially used #[serde(tag = "kind")] on ColumnRef/LiteralValue/PartitionKeys/Predicate, but serde rejects internally-tagged newtype variants containing primitives. Switched to externally-tagged #[serde(rename_all = "snake_case")]. Caught by a_variant_serde_roundtrip_filter.

Build + test

• cargo build --release -p controller — clean
• cargo build --release -p query_engine_rust — clean
• cargo test -p controller --lib — 666 passed (was 655 after Batch 1; +11 new tests covering schema pass-through + Predicate translation + serde round-trip)
• cargo test -p query_engine_rust --lib -- engines::warm_tier — 13/13 pass

Diff: 8 files, +794 / -8

🤖 Generated with Claude Code