Confusion noise sensitivity curve change

[TomImpens]

As suggested in #125, implemented the Karnesis confusion noisein psd.py; changed this to default in all functions calling the LISA psd instead of Robson+19. Changed the documentation accordingly.

Wasn't sure what column of Table II to use (the exact confusion noise depends a bit on the iterative subtraction / smoothing scheme employed), but all give very similar curves.

[TomImpens]

Ah - missed the tests. First time using github on a public project :). I'll have a look at the failed assertion test on the mission observation time. I think the rest is a side effect of the np.trapz deprecation if I understand correctly, didn't change that in this pull request and just downgraded numpy on my side to make sure everything runs as I wasn't sure of commit courtesy w/ several issues.

[TomImpens]

I believe the test behavior in test_mission_length_effect() to be incorrect.
It fails on the case where noise = loooong.
In this case, above = noise (10 year) > regular (4.5 yr), which is false everywhere. The test is considered to fail when above is anywhere false. I believe this makes sense for 'noise' corresponding to durations below 4.5 years (on these no error is raised), but does not for the 10 year loooong case.

(I don't know whether I can or should modify tests in pull requests however, so I will not resubmit for now - on my end the commit works and gives expected behavior provided numpy is downgraded to avert the trapz issue)