docs(query-engine): describe query DAG node inputs

Author: milindsrivastava1997

asap-query-engine/src/engines/query_plan/README.md

@@ -5,12 +5,13 @@ debug-only for now: the existing executor remains the source of execution.
 
 Nodes are connected by `n<id>` inputs:
 
-- `StoreRead` fetches one aggregation over its requested timestamp bounds.
-- `ComposeWindows` turns stored buckets into one aggregate per output timestamp.
-- `ResolveKeys` combines a value branch with an optional separate keys branch.
-  Without a keys branch, the value accumulator supplies its own keys.
-- `Estimate` queries the accumulator statistic with its parameters.
-- `LimitTopK` and `Format` are presentation nodes.
+- `StoreRead` — input: none; fetches one aggregation over its timestamp bounds.
+- `ComposeWindows` — input: one `StoreRead`; composes buckets for each output timestamp.
+- `ResolveKeys` — inputs: a value composition and optional keys composition; resolves output keys.
+  Without a keys input, the value accumulator supplies its own keys.
+- `Estimate` — input: `ResolveKeys`; computes the requested statistic with its parameters.
+- `LimitTopK` — input: `Estimate`; keeps the highest-ranked `k` candidates.
+- `Format` — input: `Estimate` or `LimitTopK`; applies protocol-specific output labels.
 
 `StoreRead` is either a tumbling grid scan or a sliding exact-cover scan. A
 range plan has one read branch shared across all output timestamps. Binary