Follow-up on the run_metadata changes

[bcdurak]

Describe changes

This PR optimizes the way that we store run metadata related to different entities. (addresses the review comment here)

In our old implementation, when someone calls log_metadata, it is possible that they attach the same metadata to different entities such as pipeline runs, step runs, and model versions. This process used to create X different entries with the same key-value pair in the metadata table.

In order to optimize this process, this PR separates the previous run metadata table into two tables:

• One that holds the actual key-value pair.
• One that links this pair to different resources.

For this to work, I also implemented a new RunMetadataRequest model, that can hold more than one entity per key-value pair. In all the previous calls that used this request model, all affected models and schemas were adjusted accordingly.

Other related changes

• Preview for the docs changes
• Starter Template PR
• E2E Batch Template PR
• NLP Template PR
• LLM Finetuning Template PR

Remaining TODOs

• Add more tests (model_version_id logging)

Pre-requisites

Please ensure you have done the following:

• I have read the CONTRIBUTING.md document.
• If my change requires a change to docs, I have updated the documentation accordingly.
• I have added tests to cover my changes.
• I have based my new branch on develop and the open PR is targeting develop. If your branch wasn't based on develop read Contribution guide on rebasing branch to develop.
• If my changes require changes to the dashboard, these changes are communicated/requested.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)
• Other (add details above)

[github-actions]

NLP template updates in examples/e2e_nlp have been pushed.

[avishniakov]

I left some comments, nothing major, mostly around new flag for artifacts. If you consider this as not critical or not relevant, I will put my approval upfront to prevent PR being blocked from merging.

[avishniakov]

Why do those 2 come together here?

Suggested change

	infer_artifact: bool = False,
	artifact_name: Optional[str] = None,
	infer_artifact: bool = False,

[bcdurak]

You can potentially have a step with multiple outputs, you name one of them through the artifact_name and expect the infer_artifact to look for it in the step execution.

[avishniakov]

Why infer_artifact=True? I might be misreading this, but you gave artifact name already, no?

[bcdurak]

I also discussed with @schustmi. Eventually, we do not want to enable people to provide just a name to log_metadata for the latest version (with models or artifacts). So, if you provide just a name, it needs to be paired with infer_artifact=True, so we look for it in the output space.

[schustmi]

I think the whole function could be a lot shorter if we just store the RunMetadataResource.id and RunMetadataResource.type and do one client call in the end?

[bcdurak]

Makes sense. I trimmed it a bit already but kept the resources parameter as a list and would rather not remove it right now. I am still a bit in between if we will ever tag multiple resources at the same time and if it is alright with you, I would keep it like this.

[hyperlint-ai]

The style guide flagged several spelling errors that seemed like false positives. We skipped posting inline suggestions for the following words:

• Hyperparameter