Fix column ordering in Python, null handling for computed columns

[sc1f]

This PR provides several fixes for issues with PerspectivePython:

• When passing in a pandas.DataFrame, Perspective will read the columns property and load the dataframe in the specified order. Perspective will continue to call .keys() for datasets composed of Python dict and list values, and the canonical way to determine column order should be through creating a View with the columns kwarg set. This change simply makes Perspective's column layout more symmetrical with that of the provided DataFrame.
• When creating a View, columns included in aggregates but not in columns will not have their aggregates applied. This fixes issue Pivoted PerspectiveWidget column order broken in Python 2 #904, which occurred because the order of .keys() in aggregates !=the order ofcolumns`.
• Computed columns will write null to the output column if any value provided to the computed function is null or undefined.

[texodus]

Thanks for the PR! Looks good!

At some point we should return and specify the semantics of "natural column order" of a perspective.Table