Possible updates to combiner class

Okay so, @eteq recommended I run this stuff by you guys before I started on anything. I've looked through the current stuff in the ``Combiner`` class and it looks like there's two things going on.

1) the combinations are being done with numpy masked array arithmetic calls. If these were changed to use the NDData arithmetic it would take care of the uncertainty arrays automatically. Are you guys okay with me changing these calls?

2) in addition to that, the "use the provided uncertainty" arrays when calculating the final uncertainty needs to be hooked into the actual ``average_combine``,``sum_combine``, ``median_combine`` methods one way or another. I'm not sure if this also involves some tweaks to the ``combine`` function.

To get what I need out of ``Combiner`` I need to do 2 wether or not I do 1. Thoughts?

The issue names the Combiner class and its average_combine, sum_combine, median_combine, and combine methods; start by tracing their masked-array arithmetic and provided-uncertainty handling. Clarify the intended NDData arithmetic and uncertainty behavior, then verify that the agreed results are covered before changing the implementation.

Written by the indexing model from the issue text.