Distinguish between different starting frequencies

As detailed in #314 (closed), currently pesummary assumes a single low frequency cut-off for all detectors (when evaluating the likelihood integral), and for waveform generation. It would be nice to store different frequencies, similar to how bilby_pipe performs ({H1:20, L1:15, waveform:10}).