Hatchling bypass-selection lets %pip install a deps-only pyproject

For a Databricks repo with no Python package to ship (only notebooks), you can still keep dependencies in pyproject.toml and have the notebook run %pip install -q .. from a Git folder or bundle deployment: set [tool.hatch.build.targets.wheel] bypass-selection = true so hatchling builds an empty wheel instead of failing with 'Unable to determine which files to ship'. Verified on Free Edition serverless via databricks bundle run. Also: deleting a Unity Catalog registered model fails with 'Function ... is not empty' until you delete its aliases and every model version first (databricks registered-models delete-alias, databricks model-versions delete).