Improve calc_precision_recall function speed

[theodumont]

As suggested in #1, this PR improves calc_precision_recall speed using numpy. Fix #1.

Minor additions to #1 version

• added a try/except to catch ZeroDivisionError
• use np.any(d < threshold*threshold) to produce exact same output as original function

Benchmark

A quick benchmark to see how it improved speed

from time import time
t0 = time()
for _ in range(1000):
    calc_precision_recall(contours_gt, contours_pr, threshold)
print(f"Executed in {time()-t0:.2f}s.")

gives

# original function
Executed in 60.53s.
# with numpy
Executed in 5.45s.

Thanks!