feat(planner): --planner saved reads RQE trees and writes rqe_configs #882

Part of the RQE frontend tracker. Decisions live in the tracker. Stacked on A1.

What

asap-planner-rs --planner saved (src/optimizer/saved_plan.rs):

1. When an e2e query has expr, take its leaves (and their RQE ids) from the tree instead of decompose_to_leaves on the PromQL/SQL text. Queries with promql/sql and no expr keep working as today.
2. Write a new rqe_configs section to inference_config.yaml: rqe_id -> aggregation refs (the same references query_configs uses, i.e. enough for the engine to build an AggregationIdInfo: key + value aggregation ids/types), including any query-time roll-up the deployment needs (deployment grouping finer than the RQE's).
3. Keep writing query_configs for queries that have promql/sql.
4. Engine side: InferenceConfig deserialization accepts rqe_configs (stored, not yet used).

Done when

• unit tests: an expr-only query is planned; rqe_configs is written for plain and roll-up deployments; old tables convert unchanged