Numpy errors are supressed by default

Some coordinate systems can not represent some vectors. For example, if adding two vectors in pt,phi,eta,E coordinates that have opposite azimuthal components that add to zero, the resulting longitudinal coordinate eta will be infinity:

import numpy as np
import vector
v1 = vector.obj(pt=1, phi=0,     eta=1, E=1)
v2 = vector.obj(pt=1, phi=np.pi, eta=1, E=1)
new_vector =  v1 + v2  # this vector can not be represented in the pt,phi,eta,E coordinate system

By default, such errors are suppressed in the library
https://github.com/scikit-hep/vector/blob/69d0b1d04f268540b2e553a927a8b51a7edb7633/src/vector/_compute/lorentz/add.py#L201
Without that line, the default Numpy error would be shown

/usr/lib/python3.10/site-packages/vector/_compute/spatial/eta.py:43: RuntimeWarning: divide by zero encountered in arctanh
  return lib.nan_to_num(lib.arctanh(z / lib.sqrt(rho ** 2 + z ** 2)), nan=0.0)
vector.obj(pt=1.2246467991473532e-16, phi=1.5707963267948966, eta=1.7976931348623157e+308, E=2)

I think it would be better if the user is made aware of the error by default (rather than suppressing by default). Additionally, that leaves the choice of suppressing errors or not to the user, by adding numpy.errstate(all="ignore") in their code.

Start by reproducing the coordinate-addition example in the issue, then inspect the suppression logic at src/vector/_compute/lorentz/add.py around line 201. Confirm the default behavior and decide how the existing numpy.errstate guidance should apply; done means the default operation exposes the numerical warning without suppressing it.