`acosh` returns wrong result

[antonwolfy]

The following example does not match NumPy and oneMKL:

import dpnp, dpnp.tensor as dpt, numpy as np

dpnp.__version__
# Out: '0.21.0dev0+36.g6b9774b489a'

a = dpt.asarray(complex(-4.45712982e+08, 1.0))
dpt.acosh(a)
# Out: usm_ndarray(inf+0.j)

a = np.asarray(complex(-4.45712982e+08, 1.0))
np.acosh(a)
# Out: np.complex128(20.608332945268156+3.141592651346197j)

a = dpnp.asarray([complex(-4.45712982e+08, 1.0)])

dpnp.acosh(a)  # oneMKL VM is used for 1D input array
# Out: array([20.60833295+3.14159265j])