Taylor Diagram class is not very robust

[jukent]

Things that have been giving me trouble when working on the Taylor Diagram example:

• need taylor.add_xgrid(np.array([0.6, 0.9])) to enforce X-Y ratio. This shouldn't be necessary. User should have the right ratio without needing to specify diagonal lines from origin to correlation
• dia.add_model_set(stddev=[0.6], corrcoef=[0.24]) says it accepts floats, but I need to put the values in a list for it to not raise an error.
• Should add_xgrid() and add_ygrid()be renamed to add_corr_grid() and add_stddev_grid() since the x and y nature of this plot isn't that intuitive?
• How to treat negative correlations? Add optional second quadrant to the left.
• Changing figure size messes with legend locations - add note that best at 12,12
• Add arrow annotation support to draw directional association between same variable of different models.

[jukent]

Added a small note on the figure size to the doc string in #219.

Negative correlations in lowest priority.
Needing enforce X-Y ratio is higher priority, but not well understood
Renaming grids needs discussion on best path forward
Arrows is also low priority, but higher than the negative correlation