any ground truth boxes.

processing page.

objects.

object from the structural_object_detection_trainer to the
structural_svm_object_detection_problem.  This allows us to use image scanners
which require an image be loaded before get_best_matching_rect() can be called.
I also made it so that the scanner loading (and therefore feature extraction)
is threaded.  Previously, it only used a single core.

test_box_overlap object.  However, one place in the code assumed the
test_box_overlap object was used and so this feature has actually been broken
for some time.  Moreover, it's probably just confusing and excessively complex
to have this level of user modifiability so I removed the option and hard coded
everything to use the test_box_overlap object.  This makes the code
significantly clearer.

methods to not print updates after the first print.

print the final newline if any of the verbose messages printed.

a bool indicating if it actually printed anything.