Supported datetimes with microseconds, and those with long time series (>160 years).

[pelson]

This pull request updates the AutoDateLocator and AutoDateFormatter classes to handle extreme time series (<1 second and >160 years).

The follow code demonstrates the problems:

import datetime as dt
import numpy as np
import matplotlib.pyplot as plt


d1 = dt.datetime(1990, 1, 1)

if False:
    n = 365 * 200
    x = np.array([d1 + dt.timedelta(days=i) for i in range(n)], dtype=object)
else:
    n = int(1e4)
    x = np.array([d1 + dt.timedelta(microseconds=i * 10) for i in range(n)], dtype=object)

y = np.sin(np.linspace(0, np.pi * 2, n))

plt.figure(figsize=(12, 6))
plt.plot(x, y)

ax = plt.gca()
dt_loc = ax.xaxis.get_major_locator()
plt.show()

(try running this before this PR to see the improvement)

[NelleV]

Small typo: nonsigular => nonsingular

[pelson]

Thanks @NelleV - I've added a commit which addresses your comments.

[NelleV]

LGTM 👍 for merge !

[efiring]

typo: singular

[pelson]

Thanks @efiring - I've done all of the above. The dates module is also now completely pep8 compliant.
I've also removed the pdf and svg testing, which was not adding any value to the test suite (other than testing the svg and pdf backends which are being tested elsewhere).

I think this is good to go.

[mdboom]

There are some Travis failures here, but once those are resolved, I think this looks good.

[pelson]

I've updated the failing test result (and removed the formats which are not exercising anything new).

[pelson]

Again, I think this is good for merge.