Update long running stress tests and add actor death test.

[robertnishihara]

This converts @stephanie-wang's actor fault tolerance test to a long running workload. This also changes the instance type to m5.xlarge (4 vCPUs).

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/12573/
Test FAILed.

[ericl]

Do the rllib tests still run? I think they are configured assuming a certain number of cores.

[robertnishihara]

@ericl yep, the tests all still seem to run. I've been running all 7 of them for 5 hours and the RLlib/Tune ones are still going strong (on m5.xlarge).

$ /Users/rnishihara/ray_scratch/ray/ci/long_running_tests/check_workloads.sh --load
actor_deaths: To run a command as administrator (user "root"), use "sudo <command>".
See "man sudo_root" for details.

 04:35:00 up  5:28,  2 users,  load average: 13.99, 13.52, 13.08
apex: To run a command as administrator (user "root"), use "sudo <command>".
See "man sudo_root" for details.

 04:35:05 up  5:28,  2 users,  load average: 23.47, 22.92, 22.46
impala: To run a command as administrator (user "root"), use "sudo <command>".
See "man sudo_root" for details.

 04:35:07 up  5:28,  2 users,  load average: 11.12, 12.12, 12.16
many_actor_tasks: To run a command as administrator (user "root"), use "sudo <command>".
See "man sudo_root" for details.

 04:35:09 up  5:28,  2 users,  load average: 0.00, 0.00, 0.00
many_tasks: To run a command as administrator (user "root"), use "sudo <command>".
See "man sudo_root" for details.

 04:35:12 up  5:28,  2 users,  load average: 0.00, 0.00, 0.00
node_failures: To run a command as administrator (user "root"), use "sudo <command>".
See "man sudo_root" for details.

 04:35:14 up  5:28,  2 users,  load average: 0.00, 0.00, 0.00
pbt: To run a command as administrator (user "root"), use "sudo <command>".
See "man sudo_root" for details.

 04:35:16 up  5:28,  2 users,  load average: 10.56, 12.09, 11.62

[ericl]

Looks good then, this is a nice cost savings.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/12592/
Test FAILed.