[ML] Check for model deployment in inference endpoints before stopping

[davidkyle]

It is possible to use the ml stop trained model API to stop a deployment used by an inference endpoint, this breaks the endpoint and it can no longer be used for inference. This PR adds a check to the ml stop API (_ml/trained_models/x/deployment/_stop) and rejects the request if deployment is used by or managed by an inference endpoint. The force options overrides this check.

If an inference endpoint is broken because it's model has been stopped then the to fix it is to redeploy the model using the inference endpoint Id as the deployment Id.

There are 2 cases to check for:

• The deployment was created by the inference endpoint
• The inference endpoint was created by using an existing deployment

In both case the stop deployment request will fail unless the force parameter is used.

Closes #128549

[elasticsearchmachine]

Pinging @elastic/ml-core (Team:ML)

[elasticsearchmachine]

Hi @davidkyle, I've created a changelog YAML for you.