Pin mlflow version in credit_defaults_model to fix deployment liveness probe 502

[Copilot]

deploy-model.ipynb fails at the ManagedOnlineDeployment cell with BadArgument: Liveness probe failed: HTTP probe failed with statuscode: 502, because the container crashes on startup during no-code MLflow deployment.

Root cause

credit_defaults_model's conda.yaml/requirements.txt pin Python to 3.8.15 and scikit-learn to 0.24.2 but leave mlflow unpinned. Azure ML auto-generates the deployment environment from these files at deploy time, so pip resolves the newest mlflow release (3.x), which requires Python ≥3.10 — incompatible with the pinned 3.8.15 runtime. The resulting broken environment crashes the inference server, which fails the liveness probe.

Fix

• tutorials/get-started-notebooks/deploy/credit_defaults_model/conda.yaml — pin mlflow==1.26.1
• tutorials/get-started-notebooks/deploy/credit_defaults_model/requirements.txt — pin mlflow==1.26.1

Version chosen to match mlflow_version: 1.26.1 already recorded in the model's MLmodel file, keeping the generated serving environment consistent with the environment the model was actually serialized in.

# conda.yaml
- pip:
  - mlflow==1.26.1   # was: mlflow (unpinned)
  - cloudpickle==2.2.0
  ...

No notebook cells were modified — only the model's environment spec files changed.