validation of pivot in quiver

[tacaswell]

Noted as part of #3907, the inputs on pivot for quiver is very permissive and will take 'mid' or 'middle', however it will also work with out exception with any string input (by just silently falling back to 'tail'):

X, Y = np.meshgrid(np.linspace(-np.pi, np.pi, 16), np.linspace(-np.pi, np.pi, 16))
v = np.sin(X)
u = np.cos(Y)
plt.quiver(X, Y, u, v)
plt.quiver(X, Y, u, v, color='b', pivot='ardvaark', alpha=.5)

I propose that a) we validate the input to be in {'tip', 'tail', 'mid', 'middle'} and accept but warn anything else down to tail for one version before raising.

[efiring]

The present situation is the result of my minimalist approach when I wrote it. In the history of mpl, strict validation of inputs has been inconsistent, to say the least.
For 1.5, I think that your proposed "accept but warn" is overkill; do we really need to warn every time we tighten the validation a bit to match the longstanding documentation and reasonable expectations? It's a question of a tradeoff between safety on one side and effort/clutter on the other. I think that in this case, it is OK to keep it simple: add the validation and note it in the changelog. No warning, no test.

[tacaswell]

Fair enough.

This is all sort of balled up with how we auto-fleshout the docstrings as well.

I am working on a proposal to piggy back on the mechanism that function annotations use in pythen3 (http://python-future.org/func_annotations.html) to put the information used for validation and for building the docstrings in exactly one place (and my working on I need to figure out if it will work).