[ML] Optimize Trained Models list fetching

[darnautov]

Currently to retrieve a list of Trained Model the UI performs several API calls:

• Get trained models with associated pipelines (/ml/trained_models?with_pipelines=true&with_indices=true)
• Get trained models stats (/ml/trained_models/_stats)
• Get a list of curated trained models available for download (/ml/trained_models/model_downloads)
• Optionally fetches the download state, e.g. /ml/trained_models/<model_id>?include=definition_status

And after https://github.com/elastic/kibana/issues/175128 is implemented:

• Get list of models referenced by the inference service (/_inference/_all)

To improve the loading time we should consolidate all these calls into a single Kibana endpoint.

[elasticmachine]

Pinging @elastic/ml-ui (:ml)