Fix monitoring tests

[leahecole]

fixes #2792

[leahecole]

Still working on figuring out cleanup - realizing my initial strategy is not valid for parallelization - I need to take a break and look again later

[leahecole]

Ugh the alerts tests are failing because of our parallel test runs, and the cloud-client failure is flaky and is marked as flaky, but still being considered a failure. I'm a little stuck here and taking another break (maybe a really long one) and if other folks have input please let me know!

[leahecole]

(just reverted to a state where more tests were passing than failing) - These tests do pass locally but just don't work when run in parallel - I'm out of clever ideas for now

[tmatsuo]

@leahecole FYI the alert-client tests will be somewhat fixed with #3285

although the solution there won't be the fundamental fix.

If your problem is seeing errors like "409 Too many concurrent edits", the only fundamental solution is to use multiple projects. One option is to create a new temporary project during setup and delete it on teardown, but I feel it's too wasteful. Other option is to pre-create multiple projects and pick one of the projects randomly.

Or, even we can completely avoid the problem by creating project with names following the build name.
For example, python3.5 will use the project python-docs-samples-test-python3.5 etc etc

[leahecole]

Closing this because I think @tmatsuo is looking into this, and I'd lost track of it completely