Manual import script should emit mlflow_run_id and mlflow_experiment_id columns #107

Summary

The manual import script (manual_runs/scripts/vllm/import_manual_runs_json_v2.py) does not emit mlflow_run_id or mlflow_experiment_id columns. These were added to the dashboard CSV schema in PR #83 to render clickable MLflow artifact links in the Filtered Data table.

When a manually-imported CSV is appended to the consolidated S3 CSV, rows imported via the script have empty values for these two columns and therefore show no MLflow link in the Filtered Data table — even though the source data came from a known MLflow run.

Current behaviour

• Script outputs 53 columns (as of PR feat(manual-import): add turn_index column to manual run CSV output #105/[staging] Manual import: generate per-turn rows from multi-turn benchmarks #106)
• S3 consolidated CSV has 55 columns — mlflow_run_id and mlflow_experiment_id are populated by the Forge pipeline but absent from the import script output
• Dashboard's MLflow link column (added in PR Update overview defaults to 3.5-GA, add MLflow links, remove custom Pareto section, and fix cross-version comparison #83) renders None for manually-imported rows

Expected behaviour

The import script should accept --mlflow-run-id and --mlflow-experiment-id CLI flags (both optional, default to empty) and include them in the output CSV. When a user imports a run they fetched from MLflow, they can pass the run ID so the dashboard shows a working MLflow link.

Proposed fix

Add two optional CLI arguments:

--mlflow-run-id       MLflow run UUID (32-char hex), e.g. c6aa48a0d312448380621e1bf00a8a5d
--mlflow-experiment-id  MLflow experiment ID, e.g. 264

Include them in the fieldnames list and populate the row dict accordingly.

Context

• PR Update overview defaults to 3.5-GA, add MLflow links, remove custom Pareto section, and fix cross-version comparison #83 added mlflow_run_id / mlflow_experiment_id to the dashboard schema
• PR feat(manual-import): add turn_index column to manual run CSV output #105 / [staging] Manual import: generate per-turn rows from multi-turn benchmarks #106 added per-turn row generation to the import script
• Validated end-to-end: import script output vs pipeline dashboard.csv for run c6aa48a0d312448380621e1bf00a8a5d matched exactly on all metrics — only mlflow_run_id / mlflow_experiment_id were missing