feat(planner): support nested temporal aggregation pipelines #772

Goal

Support native PromQL execution for nested aggregation pipelines whose innermost expression is a temporal function over a range selector.

Motivation

The differential compliance suite can currently cover one temporal function plus one outer aggregation, but the planner rejects a second aggregation layer such as topk(3, sum by (job) (rate(data[5m]))). The compliance runner disables forwarding for unsupported queries, so those expressions cannot be exercised as native Prometheus-vs-ASAPQuery comparisons.

Scope

• Support nested aggregation pipelines with a temporal anchor, including rate, increase, sum_over_time, count_over_time, avg_over_time, min_over_time, and max_over_time where their statistics can be represented correctly.
• Support inner aggregation grouping with no grouping, by (...), and without (...).
• Support outer reducers (sum, count, avg, min, max) and topk where their composition is semantically valid.
• Preserve PromQL instant/range semantics and TopK ordering: instant TopK values are descending within each group; range matrices have no ordering requirement.

Acceptance criteria

• The planner generates native configurations for representative nested temporal queries such as topk(3, sum by (job) (rate(data[5m]))) and avg(sum by (job, instance) (rate(data[5m]))).
• Native /query and /query_range results match Prometheus at shared evaluation times.
• Grouped TopK preserves within-bucket order for instant queries without imposing a bucket order.
• Unsupported compositions fail explicitly rather than silently falling back or dropping an aggregation stage.

Follow-up testing

Once this capability exists, add the full temporal-anchor × inner grouping × outer-reducer compliance matrix.