Fixes ray dashboard hanging problem (#1088)

[Michaelvll]

Closes #1088.

This fixes the ray dashboard hanging problem. By checking the py-spy dump --locals --pid <dashboard.py pid>, we found that the dashboard has some leaked thread _monitor_job in the dashboard/modules/job/job_manager.py. The problem is caused by await job_supervisor.ping.remote(), which may not normally raise the exception after the actor job_supervisor is exited.
After switching the await to ray.get, it seems the problem is solved.

Tested:

• Run the reproduce code in the issue description and the problem does not occur.
• Tried on the user's program and it did not stuck after submitting 500 spot jobs during the night. (Previously, it would stuck after 100-200 jobs)

[concretevitamin]

This is perhaps the toughest bug to track down so far @Michaelvll! Great debugging.

Consider running the smoke tests before merging, as it's deep in the job submission code path.

[Michaelvll]

Good point!
Tested:

• tests/run_smoke_tests.sh

[Michaelvll]

The target user's program now successfully submitted 2000 jobs. The problem should be solved. ; )