ENH: replace uses of np.isscalar with pd.lib.isscalar

[immerrr]

np.isscalar doesn't detect a lot of types that are scalar from pandas point of view, such as dates, times, None, pd.Period, so pd.lib.isscalar should be preferred instead.

Adding the issue as requested in #8706.

[immerrr]

Also, this brings me back to the idea of adding a developer API document to describe known caveats (e.g. like implications of datetime being broken for numpy-1.6, it's not used anymore but it's a prime example of what I'd expect to find there) and useful functions (e.g. pd.lib.list_to_object_array) that aren't really advertised and are unknown until you run into them and look into cython sources.

[rockg]

Having developer API documentation would be a tremendous help.