[Mnj] SPLIT patch (AGAIN!)

[Eric Greisen]

It turns out that I still did not get SPLIT right.  When averaging
some channels together while outputing a multi-channel file, SPLIT
contained two errors.  The reference pixel in the output header
was in error by (BCHAN-1)/CHINC - add this to the previously written
value to get the right value.  This could cause the frequency
ratio between channels to be wrong by some amount.  This could matter
when imaging large fields with wide total bandwidths.

Another error was made beginning in 31DEC06 which pnly affects data
with multiple IFs which have different frequency increments.  Note
that this error was made when trying to fix the previous situation in
which this possibility was widely overlooked.

A MNJ or a manual patch on 31DEC06 is now available.

Eric Greisen