Make comparsion of images more reliable

[zzambers]

Current code comparing screenshots is not very reliable and does not work consistently on different OSes:

• seems like there were some changes in ImageMagic's metrics for compare. Different imagemagick/os versions show different results (see also previous PR). Also output format sometimes differ.
• generated diff images also look different on different imagemagick/os versions (some are grascale, some use red color for differences)
• for solid black background (first screenshot) import uses 1-bit depth png, while later one it uses 24-bit depth. When compared by some versions of imagemagick, it then compares only first channel (red) instead of all channels.
• values describing similarity seem very abstract/artificial, not sure how they corelate to diff image

This tries to fix above issues. I have tried to play with metrics in compare command, but they just don't behave consistently in different versions. So I switched to different approach. Compare uses explicit parameters to generate mask of differences (black and white). Difference as proportion of changed pixels in % is then computed. (this now has direct correlation to what is seen on difference mask and is easier to reason about) Also 24-bit png is explicitly specified for import, to behave more consistently. I have done testing on different rhels and this code seems to behave consistently.