refactor(types): arrange asap-types modules by #511 section (#572 B1)

[zzylol]

Rebased on main d4869a7 (DF 54).

Step B1 of Phase B of #572 (tracker #528): arrange asap-types by #511 section instead of the pre-ASAP / post-ASAP split.

Why

pre_asap and post_asap no longer describe their contents. There is one operator DAG (#511), pre_asap::schema is the schema of every DAG, and post_asap::sketch is shared vocabulary that the logical stage uses. The later Phase B crates (logical-optimizer, physical-optimizer, plan-selection, executor) should import types by what they are, not by a stage name that no longer exists.

What

Before this PR

crates/types/src/
  ir/          node, non_asap, asap, operator_properties, scalar, aggregate_schema, error,
               timing, summary_coverage, export, physical_export, wire, cse, canonicalize,
               query, schema_support
  pre_asap/    agg_intent, expr_ir, schema, scalar_type_rules, column_resolution
  post_asap/   sketch, guarantee, execution_data_state, maintained_population,
               summary_window, query_time/
  workload.rs, parsed_workload.rs, resources.rs + resources/
  cost.rs, dag_export.rs, serde_f64.rs, types.rs

After this PR

crates/types/src/
  ir/
    operator/    node, non_asap, asap, operator_properties, agg_intent, maintained_population
    scalar/      ScalarExpr (mod.rs), expr_ir items, scalar_type_rules, column_resolution
    schema/      Schema/Field/FieldDataType/DataType (mod.rs), aggregate_schema, error,
                 state_type (was post_asap::sketch)
    properties/  guarantee, timing, summary_coverage,
                 execution (ExecutionTiming / DataPrimitive / ExecutionDataState / ExecutionDataStateError)
    export, physical_export, wire, cse, canonicalize, query, schema_support
  workload/      workload (mod.rs), parsed_workload, resources/
  physical/      execution_data_state (lift_plain, ExactOperationSchemaError), summary_window
  cost.rs, dag_export.rs, serde_f64.rs, types.rs   (unchanged)

• Public type names do not change. Only module paths change.
• post_asap::query_time had no callers and is deleted, along with its 20 unit tests.
• No compatibility re-exports from the old paths. Each section's mod.rs re-exports the vocabulary that pre_asap / post_asap used to re-export, so pre_asap::AggIntent becomes ir::operator::AggIntent, post_asap::SketchAlgorithm becomes ir::schema::SketchAlgorithm, and so on. The ir root keeps its existing re-exports (OperatorNode, ASAPOp, the timing functions, …).
• lib.rs crate docs and the docs that cite module or file paths are updated.

How

• git mv for every file, so history follows the moves. Similarity is 88–100 %.
• A script parses every use tree rooted at asap_types, planner_types / planner (the alias in asap-physical-operators) or crate (inside asap-types), maps each leaf path to its new module, and regroups the result. It does the same for inline qualified paths. Then cargo check iteration and cargo fmt.
• Measured volume: 217 files changed (204 .rs). About 580 use statements rewritten, plus about 345 inline qualified paths.
• Behavior-preserving. #[test] count goes from 1793 to 1773, and the 20 removed are exactly the deleted query_time tests. No other test is changed beyond its import paths.

Gate at this tip: cargo fmt --all --check, cargo clippy --workspace --all-targets --all-features -- -D warnings, cargo test --workspace (1518 passed, 0 failed, 12 ignored) and python3 -m unittest discover -s tools/dag-viewer -p test_render.py (29 run, OK, 6 skipped) all pass.

Deviations from #572, for review

• ExecutionDataStateError and DataPrimitive go to ir::properties together with ExecutionTiming / ExecutionDataState, not to physical. The reason is the one Reorganize crates and modules by #509 stages and the #511 unified IR #572's update gives for the split: ir/timing and ir/physical_export use them. What stays in physical::execution_data_state is lift_plain and ExactOperationSchemaError. Their only callers today are Stage 1 (exact_composition), so physical may not be their final home.
• expr_ir.rs is now a private file of ir::scalar, and its four items are re-exported there. This avoids an ir::scalar::expr_ir path that keeps the legacy name.
• Not listed in Reorganize crates and modules by #509 stages and the #511 unified IR #572, so left in place: ir::schema_support (PromQL series-identity realization) stays at the ir root, and cost, dag_export, serde_f64, types stay at the crate root.
• tools/dag-viewer/post_asap_fixture.json still exists but nothing references it, so it is not renamed. PostAsapDAG / pre_asap_sub_dag are gone from code. PostAsapDAG is still mentioned in prose in docs/design_docs/concepts/post-asap-ir.md, docs/design_docs/physical-planning-and-deployment.md and docs/develop_docs/library-api.md, which this PR leaves alone.
• Migration tables in docs/develop_docs/*-migration.md keep their historical pre_asap:: paths.
• Step B0 (moving the CostModel trait and recurrence types into asap-types) is not in this stack, and this PR is based directly on test: planner-layering Examples 1–4 acceptance specs on the planning stages #577. B0 cannot be done as a types-only move:
  • The trait's signature needs Stage 1 types that carry behavior: ReplacementSubDAG, TargetSubDAG, Replacement, Realization, ExactComposition, ExactOperation and OperationPlacement. For example, ExactComposition::compose_with_accuracy uses AccuracyModel, and OperationPlacement::provenance returns ReplacementProvenance.
  • Two default method bodies call behavior: size_params calls accuracy::estimators::size_params, and cse_share_decision_with_recurrence calls recurrence::decide.
  • B0 needs a decision before it can proceed.

Refs #572 · #528

🤖 Generated with Claude Code