Production KLL, HLL, Count-Min, and CountSketch queries return No result #491

Context

#478 made the all-sketch whole-path matrix strict and removed ignored coverage. These scenarios successfully plan a workload, install the streaming configuration, ingest a modified-OTLP sketch, and reach the public query endpoint, but the warm query path returns:

{
 "status": "error",
 "error": "No result for query",
 "infos": ["data_source: asap_query"]
}

Affected scenarios:

• KLL instant: quantile_over_time(0.5, request_size_bytes[10s])
• HLL instant: count(unique_users_per_min)
• Count-Min Sketch instant: count_over_time(endpoint_request_freq[10s])
• Count-Min Sketch range: the same query through /api/v1/query_range

This is distinct from the typed-binding failures filed in ASAPPlanner and from the HLL cumulative multi-SID classification tracked by #431. Exact Prometheus range-step semantics are tracked separately by #487.

Reproduction

./scripts/e2e.sh whole-matrix

Relevant tests in data_plane/tests/e2e_controller_plans_and_backend_serves.rs:

• controller_plan_to_query_full_roundtrip_kll
• controller_plan_to_query_full_roundtrip_hll
• controller_plan_to_query_full_roundtrip_count_min_sketch
• controller_plan_to_range_query_count_over_time_cms

Triage needed

For each family, record and verify:

1. emitted aggregation type and parameters;
2. derived policy fingerprint;
3. minted SID and family;
4. candidate/capability resolution;
5. windows returned by SketchStore::query_range;
6. SummaryExecutor readout selection and decode result.

The current top-level No result response does not identify which stage dropped the candidate/data.

Done when

All four scenarios return a non-empty ASAP-local result with correct labels, result shape, and family-specific oracle assertions. If they have separate root causes, split this issue after the first-stage diagnostic identifies them.