refactor(stage2): physical expansion (decision-tree levels 3–4), physical edges, pruned alternatives; Count-Min weights via column_sign (#1207 PR 5)

Cargo.toml

@@ -4,11 +4,13 @@ members = [
     "crates/types",
     "crates/frontend-common",
     "crates/sql-function-catalog",
-    "crates/logical-asap-optimization",
-    "crates/physical-asap-optimization",
+    "crates/stage0-memo",
+    "crates/stage1-pass1-per-computation-candidates",
+    "crates/stage1-pass2-reuse-candidates",
+    "crates/stage2-physical-optimization",
     "crates/accuracy",
     "crates/cost",
-    "crates/plan-selection",
+    "crates/stage3-plan-selection",
     "crates/frontend-promql",
     "crates/frontend-metricsql",
     "crates/metricsql-common-parser-support",

crates/accuracy/Cargo.toml

@@ -3,20 +3,14 @@ name = "asap-accuracy"
 version = "0.1.0"
 edition = "2021"
 
-# The accuracy model shared by the planning stages: per-family estimators,
-# guarantee composition and propagation through a DAG, sizing, budget
-# split, the target check and the
-# deployment's accuracy model and evidence traits. Depends only on
-# `asap-types`, never on a stage crate, so every stage can use it.
+# The accuracy rules shared by the planning stages: per-algorithm
+# estimators (the theoretical guarantee of θ), the target check, the
+# deployment's measured accuracy and α^log of a logical edge. Depends only
+# on `asap-types`, never on a stage crate, so every stage can use it.
 [dependencies]
 asap_sketchlib = { workspace = true }
 asap-types = { path = "../types" }
 serde_json = "1"
 
-[features]
-# Exposes `test_support`: estimator, sizing, composition and propagation
-# helpers other crates' tests call.
-test-support = []
-
 [dev-dependencies]
 asap-types = { path = "../types", features = ["test-support"] }