`array_api_compat.torch.round` raises `NotImplementedError` for complex input

[mdhaber]

The standard defines round for all numeric input, but:

from array_api_compat import torch as xp
xp.round(xp.asarray(1.5 + 1.5j))
# Traceback (most recent call last):
#   File "E:\miniforge3\envs\scipy-dev\Lib\site-packages\IPython\core\interactiveshell.py", line 3747, in run_code
#     exec(code_obj, self.user_global_ns, self.user_ns)
#     ~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
#   File "<ipython-input-4-b58171f4cea1>", line 1, in <module>
#     xp.round(xp.asarray(1.5 + 1.5j))
#     ~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^
# NotImplementedError: "round_vml_cpu" not implemented for 'ComplexFloat'

[ev-br]

Indeed the test is skipped: https://github.com/data-apis/array-api-compat/blob/main/torch-xfails.txt#L132-L133
Is there an upstream issue for round?

[mdhaber]

Added to pytorch/pytorch#58743 (comment).

[ev-br]

Thanks. Fix in #408, could you review / double-check that it fixes your issue Matt?