fix(planner): bump rqe-optimizer to S8 and split Sum/Count/TopK capabilities

[milindsrivastava1997]

Part of #790 (session 3a). This is the step before A3: it moves the planner to sketch-bench S8 and to the new capability split.

What

• Cargo.lock: rqe-optimizer is now at sketch-bench 3b31dfb. That includes feat(rqe-optimizer): family properties, shared fixed-size sketches and DeltaSet key tracker sketch-bench#160 (S4b family properties and key tracker), 157 update docs to point to projectasaparroyo fork #163 (Sum/Count and TopK split, deny_unknown_fields) and Added quotes to asap-sketch-ingest quote to preserve case #168 (S8 plan output).

• Capability mapping (optimizer/milp.rs):

  ASAPQuery statistic	Capability
  Sum	Sum
  Count	Count
  Topk, topk_count_events = false (topk(k, sum_over_time(...)))	TopKByValue
  Topk, topk_count_events = true (topk(k, count_over_time(...)))	TopKByCount
  Topk, weighting unknown	new MilpError::TopkWeightingUnknown

  The other statistics map as before.

• solve_milp now returns only S8's MilpSolution: active deployments with retained_instance_count, and for each Raqe its deployment index and merged_instance_count. It no longer returns the candidate list and mapping.

• asap-optimizer-cli --milp also prints each deployment's retained count and key-tracker flag, and each Raqe's n.

Interface for A3 / 4a

• Use MilpSolution.deployments[i].deployment and .retained_instance_count, and MilpSolution.raqes[j].deployment and .merged_instance_count. raqes[j] lines up with MilpWorkload.raqes[j].
• The known gap from feat(planner): build rqe-optimizer Raqes and facts from the workload config #798 is closed: an avg rewrite's sum and count can no longer share one exact-sum deployment.

Test plan

• 3 new tests:
  • Count, TopKByValue and TopKByCount capability mapping.
  • Unknown topk weighting is an error.
  • avg's sum and count Raqes solve to two deployments.
• Existing solve tests now use the S8 fields.
• cargo test -p asap_planner, clippy with -D warnings, fmt; pre-commit ran workspace check, clippy and test.
• CLI smoke run on sketch-bench/out_10_6_26_1342/rqe_atomic_costs.json, scrape interval 15s:
  • Sum and Count get separate exact-sum deployments.
  • TopKByValue and TopKByCount get separate CMS heaps.
  • Retained counts and n print, e.g. max_over_time[1h] → x=240s, y=60s, retained=61, n=15.

🤖 Generated with Claude Code