Separate inputs for machine learning use cases

[wpbonelli]

Machine learning projects often require separate training/validation/test datasets. We should support this with:

• plantit.yaml config option indicating whether flows require ML treatment
• separate inputs on flow submission page

[wpbonelli]

For model versioning, should we:

• have workflow developers publish different docker image definition versions
• let users upload and specify models in the data store, to be pulled to the cluster at runtime (would need to be cached... models can be large)
• integrate with an existing platform, e.g. HuggingFace