Improve Ray serve functionality

[venkatajagannath]

Currently, when we submit a ray serve deployment, the job status will likely be in running state until it is taken down.

One way to get around that would be to set wait_for_completion=False, which would return control to Airflow to run the next task. But, there may be a scenario where the serve deployment is currently not ready but the following task needs to access it.

For example, If I want to deploy an AI model and then call it using a spark streaming application in the next task, the model might not be ready.

Things to check --

• What is the exact behavior of Ray Serve deployments when submitted through the SubmitRayJob operator?
• Should we introduce a new trigger (specifically for ray serve apps) which is called instead if the job is serve deployment?
• How can we make sure the UX remains consistent?