docs: ASAPPlanner layering MVP status report

[zzylol]

Summary

Adds docs/design_docs/proposals/planner-layering-mvp-status.md, a status report for the ASAPPlanner layering (#509) MVP as of 2026-10-05, and links it from the proposals index.

The report covers:

• What is done: how each of the four stages is implemented. Every PR in the 67-PR draft stack (feat(ir): add summary coverage metadata for time and population #567 → … → Remove the unread accuracy-evidence API #641) is listed with what it adds, grouped as:
  • unified IR (docs: define unified operators and SQL/PromQL scalar boundaries #511);
  • stage pipeline;
  • crates by stage (Reorganize crates and modules by #509 stages and the #511 unified IR #572);
  • Pass 2, Stage 3 cost, materialization and deployment inputs;
  • Stage Viewer;
  • gaps closed after the demo;
  • removal of the legacy search.
• How to read the stack: the base order from main (d4869a76, DataFusion 54), and how to try it.
• Demo results: the plan selected for each of the six docs: propose workload-wide planning, summary sharing, and materialization #509 examples (1, 2, 3a, 3b, 4a, 4b), with its cost.
• What is not done yet:
  • review items that block merging: feat(ir): add summary coverage metadata for time and population #567's description, feat(ir): add logical sub-DAG sharing and export without execution timing #537's review, feat(runtime): compile unified operator and scalar graphs #541's pending question;
  • open issues: Check declared summary coverage population against subtree filters #570, Reorganize crates and modules by #509 stages and the #511 unified IR #572, Pass 2 sharing and Stage 1 coverage parity with the retired MajorPass #580, Stage 3: evidence-driven cost (storage I/O, hand-off bytes) #615, SQL top-k ordered by an alias (ORDER BY c DESC LIMIT k) does not bind #619, Coverage lost with the legacy candidate_selection tests (follow-up of #580) #623, Exact Sum over an Int64 column is priced by Stage 3 but does not bind in the executor #626, Stage 1 whole-expression top-k heap keeps outer partition keys indexed against the absorbed aggregate #639;
  • known gaps without an issue, by stage: Stage 1/Pass 2, Stage 2, Stage 3 and accuracy, executor, and one decision still to make.

The report is documentation only. It is based on main so it can be read on its own; the PRs it describes are still drafts.

Test plan

• Every relative link resolves on main. Files that only exist in the stack (stage3-cost-model.md, the Stage Viewer guide) are named, not linked.
• The PR numbers and titles were checked against GitHub, and the gate and example numbers against a run at Remove the unread accuracy-evidence API #641's head (b517fbf2).

🤖 Generated with Claude Code