FIX: Precision on test_ccipcanode_v2.

[NiMlr]

The precision on this test seems to be set to sharp, as it keeps failing.

[NiMlr]

I did a test using the repo at this point in time. Namely, I ran test_ccipcanode_v2 400 times. It failed 9 times. Lets assume the repo had only this test. As we currently have 26 jobs running in Travis and given they are independent, to reach a 0.8 probability of a success of a travis run, we would need to have a probability of a single instance of this test succeeding of at least

Assuming we have a probability of the test succeeding of at least that, the probability of observing 9 instances or more out of 400 failing is smaller than

This suggests a more general guideline should exists for the randomized tests, especially considering that we have multiple independent randomized tests, not only this one. This fix does not live up to these standards. However, it fixes the errors for now.

[Stewori]

This time test_inverse failed. Still I will check this in now in the hope that no random test fails again. Keep fingers crossed...
Nice math btw. We should pin your thoughts as a separate issue.

[coveralls]

Coverage decreased (-0.7%) to 73.112% when pulling 8167644 on NiMlr:patch-2 into 9d2d1be on mdp-toolkit:master.