The actual procedure in making residual image is

a) FFT model
b) degrid and write it to the MODEL data column
c) read MODEL and CORRECTED data and subtract the first from the latter
d) grid the resultant in c)
e) FFT and make the residual image


Inefficiencies:

1) the model in b) and c) is unnecessarily written and re-read... from disk
2) There is an unnecessary copy in the visbuffer from correctedVis to vis at 
	d) for gridding


Change in process:

i) move the degrid process (in (b) ) in the loop of the residual vis
calulation degrid the model visibility in visBuffer chunk...hence no
need to be on disk and small enough for memory keep.

ii) grid the resultant from c) directly ...no need to copy around.

iii) Use DATA if CORRECTED is not around

iv) save MODEL column (if COLUMN exists) in the last loop only of visibility
reconciliation so that it is available for self-caling


Additional problem... process a) to e) is done in ImageSkyModel,
FTMachine, and SkyEquation classes and there are many tied in things
that  need to be disentangled before we achive (i) and (ii). For e.g
the grid used in a) is the same object as the residual image in e) and
is shared between FTMachine and ImageSkyModel. And this is used for
different  initialization methods in SkyEquation and ImageSkyModel.