[bananas] Serious bug in MS IMAGR

[Eric Greisen]

The 31DEC07 (only) version of AIPS has had since its inception a bug
causing it to handle Gaussian Clean Components incorrectly.  An
addressing problem caused the output visibilities when subtracting or
dividing to be unchanged.  This would cause IMAGR to be non-convergent
and get into an infinite loop in some cases (although mysteriously it
would choose another resolution sometimes having done nothing useful
with the current, non-point resolution).  Previous multi-resolution
models would not have been applied in CALIB, UVSUB, et al. correctly
by 31DEC07.

The corrected code is now available for MNJs.

Please keep close watch on 31DEC07.  The code underwent a major
revolution and oversights such as the current one are still possible.

Thank you,

Eric Greisen