Make time reconstruction more efficient

This moves the step where only the elements in the time array that are also in the prior are kept before the inner products are calculated, greatly reducing the number of necessary computations. It also applies the frequency mask to the inner products so that trash data or PSD frequencies that are meant to be notched don't break the fft. The changes to pymultinest are separate and I think have already been merged, so I'm not sure why they're showing up in the diff.