Updates to spectral() behavior

[CommonClimate]

Addresses #536
Summary of changes:

• the freq_method kwarg has been replaced by freq, which can be str or array. If a string, will use make_freq_vector with the specified frequency-generating method. If an array, this will be passed directly to the spectral method.

• the default frequency method for Series.spectral() is no longer 'log' ; it is now None, and the code assigns freq depending on method. (e.g. method=wwz triggers freq=log, but method = lomb_scargle triggers freq = lomb_scargle ). Of course, that can be overridden easily, and users can still force freq = log with any method if they so choose. Evenly-spaced methods (e.g. MTM) are not affected.

• make_freq_vector() now has a new argument, rayleigh_bound, which forces the lowest frequency to be the Rayleigh frequency (f_R) if set to True, which is the default. I am not passing this at the API level, because in my mind, if people are going to screw around with a custom frequency vector, they can generate it themselves (using make_freq_vector for instance) and pass it to the method as freq. We’ve historically done this through settings. I don’t mind making this an extra argument but I don’t really see the point.

• freq_vector_lomb_scargle()used to have default ofac=4, hifac=1.0. I’ve now reduced the default hifac to 0.95 to avoid the weirdness at the Nyquist frequency. Open to other suggestions.

• freq_vector_log used to assign a default fmin that was 2*f_R. I’m not sure where that came from (naturally it was not explained anywhere), and it created discrepancies with the other methods. I’ve now aligned that so fmin = f_R if None is provided to the method.

• docstrings have been updated, but I have not rebuilt the docs, so I expect some minor bugs. Similarly, the spectral PyleoTutorial will need a major overhaul, which I'm happy to do once this is approved.