[bananas] patches to SPLIT, SPLAT, FILLM

[Eric Greisen]

The 31DEC06 version has 3 new patches:

FILLM defaulted the nominal sensitivity differently in the weighting
and the scaling of the data.  This led to weights that were a factor
of 3 too high.

SPLAT and to a lesser extent SPLIT made an error in the IF frequencies
of all IFs > 1 when averaging channels to amke a smaller, but still
multi-channel output data set.  SPLAT always got it wrong, SPLIT got
it wrong when the channel increments of different IFs were different 
(e.g. opposite sidebands differ in sign).

A MNJ or standard patch operation may be done of 31DEC06 now.

The MNJ tomorrow will make these corrections for SPLIT and SPLAT
(FILLM was done a couple weeks ago).

Eric Greisen