Current UVBeam handling is wrong

[bhazelton]

The current code mixes UVBeams with coupling matrices and "responses" extracted from FHD. When we use UVBeam we should not use anything else, the stuff in FHD is all for setting up the average embedded element MWA beam, none of that applies to the MWA beam read into UVBeam (which is the true embedded element model).

While we're at it, we should also implement the Jones decomposition detailed in Nicel's memo.