ppStack needs to be optimised, principally to reduce the memory
footprint (10 or so images will get it to over 4 GB).  The proposed
strategy follows, more or less, the same as for ppMerge.

* Convolution steps:
  - For each file:
    + read entire image
    + convolve to reference PSF
    + write convolved image

* Combination step:
  - For each chunk:
    + For each file, read chunk
    + combine chunk
  - Probably have to repeat, to do the multi-pass rejection


To do:
* Need file PPSTACK.OUTPUT.CONV (image, mask, weight versions)
* Need file PPSTACK.INPUT.CONV (image, mask, weight versions)
* Check: does pmFPAfileChecks do piece-by-piece read?
* Can we reset a file, so that we can read it twice through?
* pmStack functions probably need work so that they operate on a piece
  of the image at a time.  Will there be issues with the masks in the
  second pass?
