Big rewrite of ImageCanvas to use a QGraphicsView

Let that do all the hard work of drawing and scaling. Eliminates
having to paint the selection box outside a paint event, allows
multiple highlights with different styles and them not being
erased by the selection box.

Selection area API cleaned up for consistent interpretation of
the rectangle: functions and signals using a QRect refer to
absolute image pixels, while those using a QRectF specify a
proportion of the image size.

Lots of other API cleaned up, consistently named and documented.

Update OCR result image position tracking. Get the ImageCanvas to
detect the click and perform the coordinate conversion, so that
OcrEngine doesn't have to know about its implementation.
OcrEngine::eventFilter() is now OcrEngine::slotImagePosition()
and much simpler.