`boundary` argument is inconsistent in affine transforms

[mwallerb]

In fixing affine transforms through PR #32, the boundary argument is now sometimes a keyword argument and sometimes it is a positional argument.

I think this should be consistent. I would vote for a positional argument, maybe even make it mandatory (i.e., removing the default), because picking boundary conditions is very much part of what you should think of when constructing an affine transform.