sign of complex 0+0j returns complex nan

[WarrenWeckesser]

Example:

In [3]: import array_api_strict as aa

In [4]: z = aa.zeros(1, dtype=aa.complex128)

In [5]: aa.sign(z)
/home/warren/py3.13.5/lib/python3.13/site-packages/array_api_strict/_array_object.py:970: RuntimeWarning: invalid value encountered in divide
  res = self._array.__truediv__(other._array)
Out[5]: Array([nan+nanj], dtype=array_api_strict.complex128)

The result should be Array([0+0j], ...).

Special handling of the complex case was added in #74, but that change didn't consider the input 0+0j.

[ev-br]

Indeed, array-api-strict is an odd one out:

In [20]: cupy.sign(cupy.asarray([1.0, -1.0, 0.+0j]))
Out[20]: array([ 1.+0.j, -1.+0.j,  0.+0.j])

In [21]: jnp.sign(jnp.asarray([1.0, -1.0, 0.+0j]))
Out[21]: Array([ 1.+0.j, -1.+0.j,  0.+0.j], dtype=complex64)

In [22]: torch.sgn(torch.as_tensor([1.0, -1.0, 0.+0j]))
Out[22]: tensor([ 1.+0.j, -1.+0.j,  0.+0.j])

In [23]: import array_api_strict as xp

In [24]: xp.sign(xp.asarray([1.0, -1.0, 0.+0j]))
/home/ev-br/repos/array-api-strict/array_api_strict/_array_object.py:979: RuntimeWarning: invalid value encountered in divide
  res = self._array.__truediv__(other._array)
Out[24]: Array([ 1. +0.j, -1. +0.j, nan+nanj], dtype=array_api_strict.complex128)

Fix in gh-172