Make spark connect supporting canceling job group

PySpark ML crossValidator relies on this:

proposal:
when spark client sending request, can we make it send with a "job_group_id", and in server side, for each request that with the same (user_id, job_group_id), handling request in a fixed thread