feat(calcite): push whole-table SUM down to zone-map stats

[dfa1]

What

Wires SUM into VortexAggregatePushDownRule. A whole-table SELECT SUM(col) now rewrites to a single-row Values folded from the per-zone SUM rows — no data segment decoded — joining the existing MIN/MAX/COUNT push-down (ADR 0013 §6, ADR 0018 Phase 2).

The fold goes through the new VortexTable.zoneSum(column), which delegates to reader.compute.ZoneReducer.sum (extracted in #183). Until now that fold was reachable only through the test-only VortexAggregates helper; this makes the planner a real consumer of ZoneReducer — the seam justification from #183.

Behaviour

• SUM(col) over the whole table → metadata-only Values, exact Long for integer columns, Double for floats.
• Abandons to the normal scan (correct, just not pushed) when:
  • a zone carries no usable sum — no zone map, or an overflowed zone (Math.addExact dropped it);
  • the SUM argument is a computed expression, not a bare column ref;
  • a WHERE filters the aggregate — whole-zone stats can't answer a filtered aggregate. That is the residual tier, deliberately not built here (next increment: ZoneReducer predicate support + boundary-zone streaming).

Scope

ADR 0013 stays Proposed. AVG pushes down only when Calcite's AggregateReduceFunctionsRule has already decomposed it to SUM/COUNT; an intact AVG is left for the scan.

Tests

• AggregatePushDownTest.sumRewritesToValuesFromZoneStats — SUM(volume) (I64 → exact Long) and SUM(low) (F64 → Double) rewrite to Values; values equal the ZoneReducer fold.
• AggregateRuleBranchTest — sum_withZoneStat_rewritesToValues (replaces the now-stale sum_hasNoZoneStat_abandonsRewrite, whose premise predated the per-zone SUM writer) + new sumOverComputedExpression_abandonsRewrite branch.
• Existing filteredAggregateIsNotAnsweredFromWholeTableStats still green — SUM+WHERE correctly bails.

🤖 Generated with Claude Code