Add data-plane and query-engine support for new aggregation families #787

This was generated by AI during triage.

Context

#762 introduces optimizer-facing aggregation families and their capability/SLA metadata. This issue makes those families deployable by implementing them in the data plane and query execution paths.

Scope

Implement end-to-end runtime support for the aggregation families added by #762:

• DDSketch for quantiles
• CountSketch and CountSketchWithHeap / sketchlib CSHeap for frequency and heavy-hitter queries
• UnivMon where its query semantics are defined

For each supported family:

1. Map the ASAPQuery AggregationType to the corresponding asap_sketchlib implementation and validate parameters.
2. Construct and update the state in the streaming/data-plane accumulator factory.
3. Serialize, merge, and window the state correctly, including supported subtract/delta behavior.
4. Implement query-time decoding and evaluation in the query engine.
5. Return results with the correct PromQL/SQL semantics, including top-k ordering and query weighting.
6. Add focused unit tests plus end-to-end plans that the optimizer can generate and the engine can execute.

Remove the explicit unsupported-aggregation-type error for a family only when its complete execution path is covered. Until then it remains optimizer-only and gated by --milp-allow-unsupported-sketches.

Dependencies

• Blocked by Add optimizer aggregation families and capability/SLA metadata #762 for the final aggregation-type and capability/SLA descriptor design.
• Coordinate benchmark identifiers and metric semantics with the accuracy-SLA follow-up from Optimizer: enforce accuracy_sla and latency_sla in candidate eligibility #757.

Part of #753.