Support Avg (and other multi-statistic aggregations) in query engine

[milindsrivastava1997]

Background

Avg requires two precomputed accumulators — one for Sum and one for Count — and a divide at read time. The planner already handles this correctly: both the SQL and Elastic DSL planners map `Avg` → `[Statistic::Sum, Statistic::Count]`, producing two separate `MultipleSum` aggregation configs.

The query engine does not support this yet. All three paths return `None` (Prometheus fallback) for Avg:

• Elastic (fixed in Elastic Avg: engine queries Rate but planner precomputes Sum+Count #456): was returning `Statistic::Rate` (wrong); now returns `None`
• PromQL: `get_statistics_to_compute()` returns `[Sum, Count]` for `avg`/`avg_over_time`; the `len != 1` guard in `build_promql_execution_context_tail` returns `None`
• SQL: `parse_single_statistic("avg")` returns `None` because `AggregationOperator::Avg.to_statistics()` yields two statistics

Root cause

`QueryMetadata.statistic_to_compute` is a single `Statistic`. Avg needs two lookups (Sum agg ID + Count agg ID) and a divide. `get_aggregation_id_info` also rejects query configs with two value-type aggregations ("Query config has two value-type aggregations (expected at most one)"), so even if we had the right statistic, the agg resolution would fail.

What a fix needs

1. A way for the engine to look up two value-type aggregation IDs from a query config (Sum and Count), distinguished by `aggregation_sub_type` ("sum" vs "count")
2. Execute both store queries, then divide element-wise (sum / count) to produce the final result
3. Handle count == 0 (emit NaN or skip)

This applies to all three query paths (Elastic, PromQL, SQL).