Fix CI test failures

[guoyuhong]

What do these changes do?

This PR fixes following test failures.

1. test/actor_test.py::test_actors_and_tasks_with_gpus and test/actor_test.py::test_actors_and_tasks_with_gpus_version_two. These two test cases failed since Stream logs to driver by default. #3892 . The test cases passed after I used longer waiting time. Did it take longer time to execute the task and one worker may had executed more than 1 task?
2. test/runtest.py::test_multithreading: This test often failed in Python 2 environment. I replaced random_string with _random_string. In the Python 2 environment, there would be duplicated object ids. The old random_string may have concurrent issue in multithread case, say it may generate the same random strings. Maybe we can remove it from our code.
3. object_manager_test.cc failed since Inline objects #3756 when VALGRIND is used. However, this test works fine in none-valgrind case. I increased the data size above the inline object threshold and the test passed. Maybe the owner needs to take a look at why the valgrind test fails when inline object is enabled.

Related issue number

N/A

[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/11735/
Test FAILed.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/11737/
Test PASSed.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/11739/
Test PASSed.

[raulchen]

Thanks for fixing them!