Refactor test_cache_decorator

[mschmidt87]

The test_cache_decorator test is not robust and often fails in PRs just because of slight timing differences. We should make it more robust.

@jakobj made the following suggestion in the discussion about #176 :

we should just increase the sleep_time a bit, maybe to 2s. this test fails as travis only gives us a single core and running multiple threads in it seems to slow everything down to the point where the effect of caching is not visible any more.

I have an additional question: What motivates the / 5.0 in the assert statement in this line:

hal-cgp/test/test_utils.py

Line 57 in 66a8e5b

assert (time.time() - t0) < (sleep_time / 5.0)