feat(promql): support topk over sum_over_time/count_over_time (ASAPQuery #706 parity) #502

Summary

Match ASAPQuery PR #706 by accelerating topk over sum_over_time and count_over_time in the backend-local PromQL profile, with the correct heap weighting and labels.

Compatibility target

ProjectASAP/ASAPQuery#706 added these supported shapes:

topk(1, sum_over_time(metric[5m]))
topk by (job) (3, count_over_time(metric[5m]))

The first must rank each series by its summed sample values (count_events = false / value weighting). The second must rank by the number of observations (count_events = true / count weighting). Both grouped and ungrouped forms are included. topk over rate, avg_over_time, and other temporal functions remains unsupported, matching ASAPQuery #706.

ASAPQuery-backend has generic heap-bearing sketch machinery and can compile an evidenced Top-K query, but its checked-in backend-local compatibility workload and process conformance test do not cover either #706 shape. That leaves the weighting and in-sketch key/label contract unverified at the actual profile boundary.

Scope

• Plan both shapes through canonical ASAPPlanner and backend-local physical compilation.
• Emit an explicit heap weight mode derived from the wrapped function; never default or guess.
• Preserve the heap item key labels, including by(...) partition labels.
• Execute instant and range queries from ready warm summaries.
• Fail closed when Top-K membership evidence is missing/stale or the heap's weighting is incompatible.

Acceptance criteria

• Compiler/config tests assert value weighting for sum_over_time and count weighting for count_over_time.
• Process E2E sends both query shapes through /api/v1/query and /api/v1/query_range, compares against Prometheus, and asserts data_source: warm.
• Negative tests keep topk(rate(...)) and topk(avg_over_time(...)) on exact fallback.
• No regression to existing raw-vector Top-K behavior.

Parent: #501