feat(asap-tools): let ClickHouse E2E run a frozen planner output #859

Problem\n\nexperiment_run_e2e supports controller.prebuilt_output_dir: it skips asap-planner and deploys the supplied streaming_config.yaml and inference_config.yaml. experiment_run_clickhouse.py always generates planner_input.yaml and invokes asap-planner in sketchdb mode.\n\nThis prevents a reproducible ClickHouse/ASAPQuery evaluation from executing a deployment chosen in a separate MILP planning step. It also makes planner time/configuration part of every measurement run.\n\n## Proposed change\n\nWhen controller.prebuilt_output_dir is set, mirror experiment_run_e2e:\n\n- copy and validate streaming_config.yaml and inference_config.yaml with config.copy_prebuilt_controller_output;\n- rsync that output to the controller directory;\n- do not generate planner_input.yaml or start ControllerService;\n- start query_engine_rust with the copied output as today.\n\nRetain the existing generated-planner path when the setting is unset.\n\n## Acceptance criteria\n\n- A ClickHouse sketchdb run with controller.prebuilt_output_dir does not invoke asap-planner.\n- It deploys the supplied configs and executes the SQL workload through query_engine_rust.\n- The runner fails clearly if the directory lacks either config or its inference config is not SQL.\n- A unit test covers both the prebuilt and normal planner paths.\n- Documentation includes a command/config example for a separately generated SQL MILP plan.\n\n## Evaluation relevance\n\nThis is needed to execute a frozen planner decision consistently while comparing ASAPQuery with raw ClickHouse and ClickHouse materialized views.