improving glass blur using numba

[prakass1]

This pull request is to speed up the computation of glass_blur using numba (https://numba.pydata.org/numba-doc/latest/user/5minguide.html). Since, the glass_blur had only numpy and loop based computations, I was able to move it to a separate method with @njit decorator improving the speed significantly. I also have added the requirements.txt and the setup.py is also updated and have added a demo file to test the same.

Steps done to run is:

pip install -r requirements.txt
python test_demo.py

Results obtained:

defocus_blur 0.9493465423583984
glass_blur 5.243793964385986 (Improved from about 1 minute to 5 seconds)
motion_blur 2.2798967361450195
zoom_blur 19.582147359848022

[michaelisc]

Looks great! Thanks for adding the demo script as well!