Need to add more generality over the examples for testing spot detectors

[vreuter]

The original transformation / preprocessing for DoG is what's parameterising the DoG detection function for the test_spot_detectors tests, but this should be randomised where possible (i.e., where there's no dependence of the expected output on the specific DoG transformation being applied).