[fix] shift registrar synthetic and white image test cases

The old test cases tried to check whether the accuracy was within 1 pixel, but actually checked whether it was withing 1 meter.
Therefore it passed even though the stitching might have been off.

The synthetic image used was of a black image with a white border, this then got cropped into two tiles, but there were no distinct features to stitch, therefore the test case failed after fixing the meter pixel bug.
The updated test case creates a synthetic image with randomly placed horizontal and vertical lines.

src/odemis/acq/stitching/_registrar.py

@@ -140,7 +140,7 @@ def addTile(self, tile, dependent_tiles=None):
     def getPositions(self):
         """
         returns:
-        tile_positions (list of N tuples): the adjusted position in X/Y for each tile, in the order they were added
+        tile_positions (list of N tuples): the adjusted position in X/Y for each tile, in the order they were added [m]
         dep_tile_positions (list of N tuples of K tuples of 2 floats): for each tile, it returns
         the adjusted position of each dependent tile (in the order they were passed)
         """