[Enhancement] Update deployment tutorial

[nightism]

Update and fix issues in tutorials

• jinja2 3.1 was release in Mar 2022, which causes an ImportError in azure-default

  • Related issue is https://github.com/MicrosoftDocs/azure-docs/issues/90567, and newer azure-default was released.
  • Update azure-default dependency to 1.42.0 in the inference.dockerfile.

• protobuf 3.20 and 4.21 was release in Apr - May 2022, which is incompatible with feast

  • Will cause TypeError when importing using 4.21; and there are some non backward compatible api release in 3.20.
  • Related issue in feast protobuf pinned to <3.20.0 feast-dev/feast#2484.
  • Restrict 3.10 <= protobuf < 3.20 in the inference.dockerfile.
  • Restrict 3.10 <= protobuf < 3.20 in the unit test pipeline.

• objectId was changed to id in Microsoft Graph API

  • Related documentation: https://docs.microsoft.com/cli/azure/microsoft-graph-migration.
  • Change tutorial instruction to use az ad signed-in-user show --query id -o tsv to find principal id instead.

• The default scikit-learn in AzureML may not be compatible with some features in the notebook.

  • Some issues are:
    • mlflow.sklearn.autolog() will lead to an ImportError.
    • Experiment is not outputting any models upon completion.
  • These issues can be tackled by manually installing scikit-learn==0.22.1 in the workspace.
  • Reference: https://docs.microsoft.com/en-us/azure/machine-learning/tutorial-train-deploy-notebook#install-packages

• Add a clean up block to delete unused service at the end of the tutorial

Closes: #57