docs: ASAPPlanner and ASAPQuery-backend integration architecture

[zzylol]

Summary

• Define system architecture with labeled ASAPPlanner/backend boundaries and a single authoritative post-ASAP semantic DAG.
• Compare existing integration with the target; retain derived physical runtime plans with explicit ownership and provenance.
• Explain repeated-dashboard sharing, capability/cost evidence, accuracy scope, deployment profiles, and high-level consolidation milestones.
• Add a migration delivery tracker linking Planner data-plane: honest SketchStore memory diagnostic + idle-sid eviction #356 and backend fix: share compatible Collector materializations across query roots #513–Select complete bound workloads with strict cost and capability evidence #522, their acceptance evidence, and deployment-only boundaries.
• Document the implemented provider-priced complete bound-workload comparison and successful same-semantic generation cutover without claiming production calibration, arbitrary semantic replacement, or universal platform rollout.
• Keep ProjectASAP-wide expansion outside this backend's required migration scope.

Before / after

Before: architecture and migration ownership were implicit across the Planner pin, controller and serving paths.

After: an explicit shared semantic contract, system workflow, staged migration plan and implementation/acceptance tracker distinguish supported-profile implementation from review, merge and production rollout.

Scope

Documentation only; no runtime changes or repository rename in this PR. The runtime changes are separate sequential PRs and are not merged automatically.

Validation

• git diff --check.
• Baseline inspection: backend 95131d83972bb7a07d338e2a5af925a20c15ddce, historical Planner pin cb50219c582d43f53ab77d3a595bd1ea4a9aa119.
• Runtime test results in the tracker were run on the combined implementation stack, not on this documentation branch.

[zzylol]

Review and pushed fixes — 2026-09-08

Reviewed the 15 currently open PRs and their integration boundaries. This is self-review, not independent approval. No PR was merged; all code updates used normal commits/merges on existing PR branches.

PR	Review outcome
#505	Fixed obsolete Planner API/pin, typed count/value updates, series/materialization identity, and multi-series TopK acceptance. Now based on #522.
#506	Updated to current #505; clarified that the negative matrix tests unregistered-query fallback, not missing arithmetic support. Process tests pass.
#509	Updated to current #505; compare all series at each range timestamp, including changing TopK winners. Process tests pass.
#511	Fixed legacy emitter dropping typed count updates; migrated both heap acceptance tests to registered temporal QueryPlans. Four canonical process tests pass.
#512	Updated merged Planner dependency and acceptance limits in the migration document.
#513	Reviewed compatible producer deduplication and conflicting-contract rejection; no additional finding in this pass. Covered by combined compiler/process tests.
#514	Reviewed merged-API adaptation and fail-closed unsupported operations; no additional finding in this pass.
#515	Fixed ranking that dropped duplicate algorithm candidates and could panic; pinned merged Planner #356 revision 378a754.
#516	Reviewed packed SUM/count, arithmetic grouping/source restrictions and non-additive reduction fallback; inherited #515 fixes.
#517	Reviewed publication, frame identities and failed-generation isolation; inherited fixes; distributed process test passes.
#518	Reviewed startup/live workload selection and query-scoped certificates; inherited fixes.
#519	Reviewed shared lifecycle demand and conflicting evidence rejection; inherited fixes.
#520	Reviewed actual Collector Rust update/output counters; inherited fixes; distributed process test passes.
#521	Reviewed staging cleanup/retry and concurrent same-semantic cutover; inherited fixes; distributed process test passes.
#522	Fixed quote preparation incorrectly requiring the default warm plan to compile; available exact manifests now survive unavailable warm implementations. Regression and all 29 binary tests pass.

Validation

• 656 control-plane library tests passed on the TopK-integrated tree.
• 977 data-plane library tests passed serially. Initial parallel execution failed five persistence tests; isolated/serial reruns passed. This is disclosed, not counted as an initial clean run.
• Latest combined fix(topk): preserve typed count updates and migrate acceptance to registered QueryPlans #511 tree: two canonical compatibility process tests and one distributed production-process test passed.
• test(promql): enforce range step parity in production E2E #509 and test(promql): add PR-derived gaps to the process E2E matrix #506 each passed their two compatibility process tests.
• Latest GitHub CI runs are tracked separately; local results are not an approval.

#511 acceptance update

User confirmed migration to the new contract. Commit 1a85748 replaces the two no-QueryPlan heap TopK tests with production backend-local tests of registered topk(3, count_over_time(top_endpoint_qps[5s])).

CMS-with-heap and CountSketch-with-heap each return all three expected identities/counts (gamma=200, zeta=150, alpha=100), with timestamps, two range steps and Remote Write retry deduplication checked. Raw values are 17 to distinguish counts from weighted sums. Unregistered instantaneous TopK follows the configured exact fallback.

All four canonical compatibility process tests passed; the remaining legacy test file compiles. This resolves the two-test contract-migration decision above, not all unrelated legacy OTLP serving acceptance. No independent approval or merge performed.