- Use CPNest ACT estimation
- Change scaling
- Add chains starting after accept > 0

This updates to an anaconda python image and removes redundant tests
rerunning examples so that the tests finish in a more reasonable amount
of time. Many requirements are also moves to an
optional_requirements.txt file.

- Moves setup to a before_script
- Remove lalsuite from requirements for CI to prevent it from failing

 Removes chainconsumer and following requirements. These where causing
the CI to be very slow. Instead, rather than plotting, just list a
warning.

This demotes CC from a requirement to something optional. If
`ChainConsumer` is called when it is not installed, a warning message is
raised.

Removes use of pickle in favour of h5 format. Uses deepdish
(https://github.com/uchicago-cs/deepdish) over the more standard
h5py for the following reasons

- h5py is built around saving 'arrays' while deepdish around saving
  dictionaries
- provides useful tools to check what data is in the h5 file
- is well written, provides good documentation and no strange error
  messages (personal experience with trying to implement h5py).

nestle is used in the tests, so is required for CI. Its lightweight and
pip installable so doesn't add too much overhead.

- Test for simulated data
- Test likelihood evaluation
- Add gitlab-ci.yml
- Remove lal and lalsim which aren't directly pip installable
- Add coverage and pipeline status to README
- Adds coverage report
- Update README.md with instructions on testing and coverage