The case where it broke was a bitmap that is very wide (3000 pixels), but not
too tall - and the small height satisfied the aPosAry.mnDestHeight >
3*mnOutHeight condition.

Such workarounds that are triggered only in very rare cases are a bad idea in
general, unless there is a unit test.  I've tested with the document from
i#75264, and it works for me - there were some improvements in the bitmap
scaling, hopefully the root cause is fixed by that, or maybe I just have more
memory; who knows.

Change-Id: I4b7a749378ef35469db78dfc680fc77f87bc4382