Sikuli is broken with Java >= 9 and its Debian maintainer has given
up. So we need something else.

Here we implement something that is almost a drop-in-replacement for
the Sikuli features we use (although the interface for keyboard
interaction has changed, resulting in tons of small changes under
features/step_definitions in this commit). It uses OpenCV for (fuzzy)
image matching, and xdotool for keyboard/mouse interaction. We didn't
bother with most of the Sikuli class hierarchy, settling with just
equivalents of Screen and Match -- we drop Pattern since we don't do
OCR but just image matching; we drop Region because so far we have
always operated on the whole screen (if we care about a subsection we
can just use findAll() and check which matches fulfills the
constraints).

This commit removes some functionality: both --retry-find and
--fuzzy-image-matching are gone. These were mainly used for "image
bumping", and I do have plans for in improved feature replacing these.

Refs: #15460