Fix hist for log=True and histtype='step'

[dmcdougall]

Rebased version of #1029 to target v1.2.x. This aims to fix #196 but is currently incomplete.

[dmcdougall]

Over in #1029 there was discussion of using the log's actual base rather than the magical number 0.1. This would be the correct solution aside from the fact that the axis scale is set to base ten inside hist (it's the default). If the user then changes the [xy]scale outside of hist, how do we change the plot to reflect the change of base without calling hist again?

[WeatherGod]

I wouldn't worry too much about a situation where the log base changes
after the call to hist. A lot of stuff breaks if you do any sort of
rescaling after the call to hist, and if it isn't already, it should be
well documented to establish your scales prior to any plotting.

[dmcdougall]

@WeatherGod Alright, thanks for that information.

Are there any other concerns that need to be addressed before this is merged? All tests pass locally on python 2.7.