Defines a Champion-vs-Candidate release gate as a Databricks Asset Bundle job that compares two `mlflow.genai.evaluate` runs and blocks the model registry alias swap on threshold failure. Use when promoting a Candidate version behind a `Champion` alias and calibrated scorers, slice metrics, and latency/cost baselines exist. Do NOT use when scorers are uncalibrated, no slice list exists, or no Champion baseline run is available.
