feat(query): dispatch sum by (...) PromQL to ExactAgg(Sum) sids (closes Option-B downstream)

[zzylol]

Summary

After PR #290's Option B unified emit, the control plane mints
ExactAgg(Sum) sids for sum PromQL candidates and the engine resolves
them correctly via instances_matching — but sum by (zone) (http_requests_total)
still returned {\"error\":\"No result for query\"} because
SketchReducer::evaluate short-circuits with UnsupportedFunction(\"sum\"):
its function_to_family dispatch table only knows sketch-backed
families (quantile / cardinality / topk / frequency). ExactAgg sids
carry Box<dyn AggregateCore> payloads (SumAccumulator /
IncreaseAccumulator / …) rather than opaque sketch bytes, so they
need a parallel reducer entry point.

What changed

• SketchStore::query_exact_agg_range(sid, t0, t1) — sister of
  query_range that yields per-(label_values_map, samples)
Box<dyn AggregateCore> payloads with the full label keys
  preserved (unlike query_precomputes_by_agg, which flattens to a
  keys-erased KeyByLabelValues).
• SketchReducer::evaluate_exact_agg(sids, agg_type, group_by_keys, t0, t1) —
  projects each sid's window state onto the requested
  group_by_keys subset, merges per-(group, window) accumulators
  via AggregateCore::merge_with, and reads Statistic::Sum for
  additive types (Sum, Increase, plus the Multiple* variants).
  MinMax is deferred — disambiguating min vs max needs the outer
  PromQL function name, which the analyzer doesn't thread through
  yet; today the analyzer doesn't produce ExactAgg(MinMax) for
  sum by, so the defensive UnsupportedCapability is fine.
• ASAPQueryEngine::execute and execute_range_promql_modern branch
  on candidate.required_capability — ExactAgg(_) → new path;
  sketch capabilities → unchanged SketchReducer::evaluate.

Smoke-test before / after

Before:

sum by (zone) (http_requests_total)
→ {\"error\":\"No result for query\"}

After (bash /mydata/mvp-smoke-test/run_smoke.sh):

sum by (zone) (http_requests_total)
→ [z0=48608, z1=48604, z2=44868, z3=44864]
   (resultType: vector, data_source: asap_query)

The B-2 / quantile / quantile_over_time queries still return the
expected per-zone DDSketch values (both instant and range forms) —
their dispatch path is unchanged (SketchReducer::evaluate for
sketch capabilities).

Known follow-up gaps (not in scope)

• topk(5, sum by (zone) (rate(http_requests_total[5m]))) — the
  outer rate-over-Sum composition is not analyzer-supported as a
  single ASAP-tier candidate today; the query falls over to archive.
  Tracked separately.
• count(http_requests_total) against the ExactAgg(Sum) sids —
  same answer-shape as sum, but the analyzer routes count to
  ExactAgg(Sum) only when paired with an inner that lowers to
  Sum; a bare count(metric) against a Sum-only metric isn't
  routed to ASAP yet.
• ExactAgg(MinMax) requires threading the outer PromQL function
  name to disambiguate min vs max; deferred.

Test plan

• cargo build -p data_plane — clean
• cargo test -p data_plane --lib — 732 passed, 2 ignored, 0 failed
  (was 727 + 2 ignored; the 5 added: 4 reducer unit tests + 1 engine
  end-to-end test for execute(&str) → ExactAgg)
• bash /mydata/mvp-smoke-test/run_smoke.sh — sum-by-zone passes,
  quantile baseline preserved

🤖 Generated with Claude Code