Enable force inference endpoint deleting for invalid models and after stopping model deployment fails

[dan-rubinstein]

Description

This change modifies the delete inference endpoint API to allow the force option to delete the endpoint even in cases where:

1. A service can not be found for the endpoint.
2. The endpoint can not be parsed when reading from the inference index.
3. Attempting to stop a model deployment for the endpoint fails.

Testing:

• Tested manually writing an endpoint with an invalid service into the inference index fails to delete with force=false but succeeds with force=true.
• Tested manually writing an endpoint with an invalid model config (ex. wrong typing for property, extra value, bad chunking settings, etc.) into the inference index fails to delete with force=false but succeeds with force=true.
• Added a line of code to force the stop model deployment call to fail. Tested that with force=false deleting an endpoint fails when model deployment can not be stopped and with force=true it succeeds leaving behind the model deployment.
• Unit tests + IT tests

[elasticsearchmachine]

Hi @dan-rubinstein, I've created a changelog YAML for you.

[elasticsearchmachine]

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

[dan-rubinstein]

@elasticmachine merge upstream