Minor tweak ylim on ecdfplot

[ericmjl]

When we plot an ECDF under sns.set_style("talk"), the default of axes ylims being (0, 1) become a tad problematic for long-tailed distributions. We get the following effect of thin lines at the upper (and I'm assuming lower as well) bounds:

An easy fix would be to set the ylim to (-0.05, 1.05). Doing so would fix the effect:

Hard-coding the ylim to just beyond (0, 1) is appropriate from a visualization perspective, I believe, because the y-axis on an ECDF, by definition, is supposed to be bound between (0, 1). This feels like an easy PR, and I'm happy to handle it 😄.

[mwaskom]

Hi, I can see how the half-clipped line when you don't have an upper spine is unattractive. But I do think it's kind of a matter of taste if you have spines. Not setting the limits and letting autoscaling handle it could be reasonable, but I'm -1 on hard-coding them at values other than 0/1 (for one thing, it wouldn't work when you have stat="count"). I also don't think I'd like what it would do in the case where you are overlaying an ecdfplot on a histogram.