Stacked hist with histtype='step' (+ new kwargs for hist) #831

I submitted this to the mailing list, but I've since realized that this is probably a better place for it. So sorry for the double-post.

I've been wanting for some time the option to make stacked histograms using the step or stepfilled histtypes. I've started working on this in a fork, but I wanted to solicit some opinions on how to change the function call.

The most straightforward thing to do would be to add two new histtypes: stepstacked and stepfilledstacked, bringing the total number of possible histtypes to 6. However, that feels kind of clunky.

I'd like to propose that instead of increasing the number of histtypes, I add a new boolean kwarg called stacked which controls whether or not the histogram gets stacked. We then don't need to increase the number of histtypes. The barstacked histtype becomes redundant, as it's identical to histtype='bar', stacked=True, but I'll leave it in for backwards compatibility, to eventually be deprecated.

While I'm at it, filling a step histogram by using a different histtype (stepfilled) seems clunky and inconsistent with how one would deal with a bar histogram. To control the fill for a bar histogram, one uses the fill kwarg, which gets passed to the patch collection. I'd propose that we use the fill kwarg in a similar way for the step histogram and eliminate the stepfilled histtype.

In sumarry, here are the changes I'd like to make, as they would be reflected in the docstring

"""
         *histtype*: [ 'bar'  | 'step' ]
           The type of histogram to draw.

         *fill*:
           If *True*, the histogram will be filled.
           If *False* only the edges of the histogram will be drawn.

         *stacked*:
           If *True*, multiple data are stacked on top of each other
           If *False* multiple data are aranged side by side if
               histtype is 'bar' or on top of each other if histtype is 'step'
"""

We can of course leave in the barstacked and stepfilled histtypes for backwards compatability.

So, are there opinions on wether this proposal seems like a good idea? I'm mostly interested in getting stacked step histograms, which I've already implemented, but it seems like I might as well clean up the function call while I'm at it. If others agree that changing the function call is a good idea, I'll be happy to implement it and include it in my pull request.