convert scale.c to scale.im so we have 8 bit/sample and double/sample

convert scale.c to scale.im so we have 8 bit/sample and double/sample
implementations of mixing scaling.

modified imtoc.perl to allow non-conditional #code sections to allow
creation ofr 8 and double/sample versions of support functions.

fixed a bug in an optimization that avoids vertically scaling when the
vertical size stays the same.

The change from double/sample only to both saved about 20% on
scalebench time (which also loads/saves the images)