Heron 0.3.1

v0.4.0 alpha 1

This alpha release provides a functional but incomplete version of the
heron inference pipeline for testing.
A number of known issues remain in the codebase, both from a code
quality perspective and from a functional perspective.

Primary among these is the sampling of gps time, which is currently
restricted to discrete bin intervals in the data space.
This will be corrected in a future version.
The gpstime sampling is performed by slicing the data vector to
produce a vector which has the same length as the generated waveform
prior to performing the residual calculation.
We can improve upon this by also evaluating the waveform appropriately
to ensure the coalescence time is correctly placed; this variation in
the waveform central time will be small, changing at most by half the
time interval of a single bin in the timeseries.

Updated beta release implementing new faster CUDA model.

Implements a provisional new API for models.

First version of the code to pass basic tests, using the pypi version of George without the HODLR operation.