[Inference API] Deleting inference endpoint if start model action returns license error

[ymao1]

Resolves https://github.com/elastic/ml-team/issues/1377

Summary

When creating an inference endpoint for a built in model, the endpoint is created first and then the model is started. Because the license is not checked until the model is started, we are left with orphan inference endpoints. This updates the logic to catch license errors returned by the PutTrainedModelAction and delete the inference endpoint that was created.

To Verify

Create an inference endpoint using

PUT _inference/sparse_embedding/my-inference-endpoint
{
  "service": "elser",
  "service_settings": {
    "num_allocations": 1,
    "num_threads": 1
  }
}

You should see a 403 license error response. Verify that the inference endpoint does not exist with

GET _inference/sparse_embedding/my-inference-endpoint

which should now return a 404