No longer need *noisy.mat files or PyPlot

[notZaki]

This pull request - which probably should've been split into two PRs - tries to speed up the installation process by making two main changes.

1. Delete the *noisy.mat files from the package itself and use a function to make the noisy data.
   The function to make the noisy data was adapted from the python scripts.
   The *noisy.mat files were deleted because they take up roughly ~70% of the package size, so deleting them should speed up the download step. But, now that I think about it, the download size probably won't change because the old *noisy.mat files will still be stored in the git history. D'oh well.

2. Allow validate() and demo() to work without needing PyPlot (issue Breaks when PyPlot not installed #6) - the functions will run but the plotting step will be skipped if PyPlot is not installed.
   This should speed up the installation step by not requiring PyPlot and it's dependencies to be installed.

3. (Unrelated minor change) When running validate(), the parameter maps now use a fixed colourscale based on the range of the true values. This ensures that the noiseless and noisy runs produce parameter maps with the same colourscale - making them easier to compare.

Aside: The CCC for QIBA 4 ve was 0.698 - slightly below the 0.70 cutoff. In order to squeeze past the cutoff, I modified the tests to round the CCC to two significant digits.

[coveralls]

Coverage increased (+1.6%) to 46.985% when pulling c890336 on notZaki:optionalPyPlot into a097a55 on davidssmith:master.

[coveralls]

Coverage increased (+1.6%) to 46.985% when pulling c890336 on notZaki:optionalPyPlot into a097a55 on davidssmith:master.

[coveralls]

Coverage increased (+1.4%) to 46.734% when pulling 6afcfdb on notZaki:optionalPyPlot into a097a55 on davidssmith:master.

[coveralls]

Coverage increased (+1.4%) to 46.734% when pulling 6afcfdb on notZaki:optionalPyPlot into a097a55 on davidssmith:master.

[davidssmith]

I like these changes. I wish I had thought to auto generate the noisy data before I published the paper. Maybe someday we can reset the git archive to make it leaner.

For now, can you squash the commits into bigger groups, like three commits based on the three bullet points above?

[coveralls]

Coverage increased (+1.4%) to 46.734% when pulling db5884b on notZaki:optionalPyPlot into a097a55 on davidssmith:master.

[notZaki]

I squashed them down to six commits.