[daip] problems with MSORT on VLBA data (fwd)

[Eric Greisen]

We have made a number of changes to MSORT - to use more memory and to
add a truely brute force sort from disk (that requires that the output
not be the same as the input).  The old algorithm (still in use when
out = in or in smaller cases) is very non-linear in its response when
the records (Npol * Nif * Nchan) are long and the file large.  I have
given it more memory to use which helps and added the other
algorithm.  The latter is faster than UVSRT (on a 9 Gbyte file with
6500 words per vis record) and needs only 2 files each 9 Gbytes rather
than 3.

If you run a MNJ on 31DEC03, you will have gotten it today or the next
time you run it.

Sorry to be so slow to fix this...

Eric Greisen