WIP: Add waveform uncertainty to likelihood calculation on GPU

This MR attempts to add support for including waveform uncertainty in the calculation of likelihoods for BBH signals.

How?

GPR models for waveforms calculate both the mean waveform and their variance. This MR includes the waveform variance as an additional noise source in the inner products between data and signal templates. The likelihood calculation is carried-out on GPU using CUDA via pytorch.

Why?

Waveform uncertainty is likely to vary considerably across the BBH parameter space, thanks to the varied availability of NR BBH simulations against which the waveform approximants are calibrated.