Floating-point precision error in calculating angles

[philipmnel]

Describe the bug
When using qcel.util.measure_coordinates to measure an angle between coordinates, an angle of 180° can cause a nan value to be returned. The bug is caused by a floating-point precision error in calculating the cosine of the angle; if the cosine is greater than 1.0 due to noise then np.arccos will return nan.

To Reproduce

import qcelemental as qcel
import numpy as np

geometry= np.asarray([[0, 0, 0], [1, 0, 0], [2, 0, 0]])
angle = qcel.util.measure_coordinates(geometry, [0,1,2], True)
print(angle)

geometry= np.asarray([[-5.53102237, -12.51893475, -8.92439275], 
                      [-5.53102237, 2.9394029, 1.97762624], 
                      [-5.53102237, 18.39774056, 12.87964524]])
angle = qcel.util.measure_coordinates(geometry, [0,1,2], True)
print(angle)

The first prints 180.0 as expected, the second returns a nan (calculated cosine is 1.0000000000000002).

Using qcelemental 0.25.0

Expected behavior
The calculated cosine should be properly bounded within [-1, 1].

Additional context

[loriab]

closed by #325