[ML] Trained models in spaces

[sophiec20]

ML jobs will be space-aware (targeted for 7.11 #64172), however trained models are not.

A data frame analytics job will create a trained model. This can be used during ingest in an inference pipeline or in search as a pipeline aggregation. https://www.elastic.co/guide/en/machine-learning/current/ml-inference.html

Trained models may also be imported into the stack and may not have a dependent data frame analytics job in the following scenarios:

a) the model could have been trained in, and imported from, another elasticsearch cluster https://www.elastic.co/guide/en/machine-learning/master/ml-trained-models.html
b) the model could have been trained in this current cluster but the job has since been deleted
c) the model could have been trained in python and then imported https://eland.readthedocs.io/en/latest/examples/introduction_to_eland_webinar.html#Machine-Learning-Demo
d) the model could have been created by a job that does not belong in the current space

Trained models can be dependent objects of analytics jobs, but they can also be first class citizens in their own right. We could treat them as management objects similar to ingest pipelines, or as space-aware objects similar to analytics jobs. Current thinking favours models to be space-aware.

[elasticmachine]

Pinging @elastic/ml-ui (:ml)