Investigate extra_keywords handling in measurement sets

[bhazelton]

The current approach is to put the entire extra_keywords dict into the main MS table as a dict called "pyuvdata_extra". Will this work if there are arrays or more complicated things present? If not we should warn about those kinds of things.

Also, this approach might not pick up generic extra keywords inserted by other MS writers. Maybe this isn't a common practice though, as it is in e.g. uvfits.