Add support for 'mlflow-artifacts' source in import

[erwin-vanduijnhoven]

This PR improves interoperability with newer MLflow Tracking Server artifact URI formats (mlflow.artifacts) and fixes several edge cases in how model artifact paths are parsed from registered model version source fields. It also adds tests to prevent regressions.

What’s changed

• Allow mlflow-artifacts: URIs as valid source values when importing model versions (skip local filesystem existence checks for this scheme, similar to dbfs:/s3:).
• Fix artifact subpath extraction by matching on artifacts/ and correcting slicing to avoid off-by-one errors that could truncate paths.
• Normalize artifact/model name extraction in copy/run utilities by consistently using artifacts/.
• Add open-source tests covering mlflow-artifacts: sources with and without an artifact subpath.

Why

• MLflow can emit mlflow-artifacts: URIs for artifact locations; imports should treat these as remote/virtual locations rather than local paths.
• Existing parsing logic could mis-handle certain source strings (missing/extra leading slash, truncated first character, ambiguous "artifacts" match).

Testing

• Added unit tests in tests/open_source/test_models.py for mlflow-artifacts: path handling.

Notes

• No behavior change for existing dbfs: and s3: sources; local file path validation remains in place for non-remote schemes.

[a-gajam]

Can you provide more description on the feature?

[erwin-vanduijnhoven]

@a-gajam could you review this PR?

[a-gajam]

LGTM