Stabalize test-suite execution

[roman]

The current version of Capataz uses concurrent tests to assert it's behaving as expected, and this has the unfortunate side-effect of the test-suite not being stable.

Ideas:

• Use deja-fu to mock schedulers -- this seems tricky, specially because we are planning to move to RIO and not certain it will work well with deja-fu
• Have a different (stable) test-suite and move existing one to a different project