The sampler no longer, by default, creates an ourput directory

1) If sigma is not given, then it will not be sampled over (or a warning
raised that you didn't specify a prior for sigma).
2) Comments added to explain the logic throughout
3) Tests added for the Gaussian likelihood

- time_duration -> duration
- fix spelling error tiime -> time
- frequencies -> frequency_array

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.

- Adds a time_array and time_domain_strain to the
  InterferometerStrainData
- Adds a test for consistency between the time and frequency domain
  strain

- Provides backward compatibility for people with scripts using the old
  style
- Also because, as Colm pointed out, there may be bugger behaviour with
  multiple detectors and it may take a long time to calculate the waveform
- Change the name in `inject_signal` from
  waveform_polarizations->injection_polarizations.

This last step could break peoples code who are using inject_signal
directly, but its probably best to unify notation now rather than have
two names for one thing.

This reverts commit 2ab4bddd.

For some reason, the list comprehension method of setting the IFOs
intereacted badly with the example_test method. This doesn't change
anything, but just makes it a straightforward for loop.

Also removes the failing detector tests.

Most of these tests fail because they rely on the old set_data method,
or try to set attributes that aren't settable.