Used NumPy functions to speed up hough_circle

[jbradt]

I changed the hough_circle function to use several weird NumPy functions instead of iterating. This made it ~100x faster (my test event changed from 754 ms to 6.73 ms).

The main differences are that it now only computes the differences between the points once, and it constructs Hrad in place rather than iterating over the thetas. The construction is done with np.repeat and np.tile. These functions (which I just discovered yesterday!) are different ways of repeating an array. So, for example,

>>> a = np.array([0, 1, 2])
>>> np.repeat(a, 3)
array([0, 0, 0, 1, 1, 1, 2, 2, 2])
>>> np.tile(a, 3)
array([0, 1, 2, 0, 1, 2, 0, 1, 2])

This appears to be much faster than looping in Python.