Allow waveform_generator to not have to be initialised with the duration and sampling frequency

This allows the waveform generator to be given None for the sampling frequency and duration (in this state, it therefore can't be used to generate an injection). Then, when passed to the likelihood, a check if performed to see if those attributes are set. If they are None, they are set from the data, and if they disagree with the data, an error is raised.

I can't decide if this is just useless bloat, or actually useful.

Closes #127 (closed)

examples/open_data_examples/GW150914.py

@@ -36,9 +36,7 @@ prior = tupak.gw.prior.BBHPriorSet(filename='GW150914.prior')
 # creates the frequency-domain strain. In this instance, we are using the
 # `lal_binary_black_hole model` source model. We also pass other parameters:
 # the waveform approximant and reference frequency.
-waveform_generator = tupak.WaveformGenerator(duration=interferometers.duration,
-                                             sampling_frequency=interferometers.sampling_frequency,
-                                             frequency_domain_source_model=tupak.gw.source.lal_binary_black_hole,
+waveform_generator = tupak.WaveformGenerator(frequency_domain_source_model=tupak.gw.source.lal_binary_black_hole,
                                              waveform_arguments={'waveform_approximant': 'IMRPhenomPv2',
                                                                  'reference_frequency': 50})