test: #509 Example 1 acceptance spec and end-to-end tests

[zzylol]

Stacked on #576 (Stage 2/3), #575, #561 and #543 (Phase C).

Why

#509 Example 1 is the end-to-end reference for planner layering. An acceptance spec written before the implementation records what each stage must produce, and which deviations remain, so a reviewer can measure the stages against the design doc and not against themselves.

What

• docs/design_docs/proposals/planner-layering-example1-acceptance.md: the MVP acceptance spec.
  • Expected counts 1 → 6 → 6 → 1.
  • Per-stage candidate tables and invariants.
  • The doc's ambiguities and the MVP deviations.
• tools/dag-viewer/examples/planner-layering-example1.expected.json: the expected-only asap-stage-pipeline/v1 shape (data only, no viewer change).
• crates/integration-tests/tests/planner_layering_example1.rs: 21 tests that run against the real Stage 0–3 (10 run, 11 ignored).
  • Tests that pass are enabled.
  • Every test that still fails is #[ignore]d, with the exact difference from the spec as the reason.
  • Runtime capability checks: the selected plan compiles in the physical planner. The check that every candidate compiles stays ignored and names the two runtime gaps.

How

The spec was written first, against todo!() stubs (first commit). The second commit replaces the stubs with thin adapters over the stages from #575 and #576. Tests identify candidates by structure, not by id.

Example 1

Before this PR: nothing checks Example 1 end to end. A stage can change its output without any test failing.

After this PR: cargo test -p asap-integration-tests --test planner_layering_example1 runs the stages and observes 1 → 24 → 24 → 1. The selected plan is P20 (Q1 exact (Sum acc, Rate acc) · Q2 exact (Sum acc)), and the test checks that it compiles in the physical planner. The ignored tests document the gaps:

• Stage 1 gives 24 candidates, not 6, because Pass 1 offers exact accumulators and CountSketch+heap.
• There is no Hydra and no shared-input variants (Pass 2).
• Only 8 of 24 physical candidates compile:
  • CountSketch+heap: the summary schema does not match what the runtime builds.
  • CMS+heap: needs non-negative weights.
• Costs are illustrative statistics.
• There is no latency check.

🤖 Generated with Claude Code