plot_waveform_posterior uses non representative samples

The following discussion from !817 (merged) should be addressed:

• @avi.vajpeyi started a discussion: (+3 comments)

  While working on this, I noticed that if n_samples is passed to plot_waveform_posterior then n samples from the top of the posterior data frame are used (the samples with the lowest LogL).

  Would it be better to use the n samples from the bottom of the data frame (the samples with the highest LogL)?

  Should I make an issue about this, or am I understanding something wrong?

I think it should take random samples. In any case, we should aim to fix this soon.