Allow xlim as parameter for lmplot

[jhncls]

Seaborn versions: latest dev version and 0.11.1

lmplot doesn't seem to accept the xlim= parameter, although FacetGrid does.

Use case: when truncate=False, the regression lines are extrapolated until they touch the current xlims. If one afterwards want to extend these xlims, the regression lines are floating again. A workaround is either to call FacetGrid and regplot separately, or to set very wide xmargins via the rcParams.

Example code.

import seaborn as sns
import matplotlib as mpl

tips = sns.load_dataset('tips')
# mpl.rcParams['axes.xmargin'] = 0.5  # set very wide margins: 50% of the actual range
g = sns.lmplot(x="total_bill", y="tip", col="smoker", data=tips, truncate=False, xlim=(0, 80))
# mpl.rcParams['axes.xmargin'] = 0.05 # set the margins back to the default
g.set(xlim=(0, 80))

[mwaskom]

What should really happen is that lmplot should accept a facet_kws dictionary that it passes to FacetGrid to set it up. Also then some of the parameters of lmplot that are passed directly should be deprecated with the instruction that they should be packaged in facet_kws (not all of them, but less-often-used ones).

Unfortunately I have not been especially consistent across the figure-level functions with which FacetGrid parameters do or do not end up i the figure-level function signature. This would probably be good to standardize, but that might involve a lot of annoying deprecation.