dpctl.tensor.logaddexp does not always return nan when one operand is not nan

[ndgrigorian]

For example:

In [1]: import dpctl.tensor as dpt

In [2]: x1, x2 = dpt.asarray(1., dtype=dpt.float64), dpt.asarray(dpt.nan, dtype=dpt.float64)

In [3]: dpt.logaddexp(x1, x2)
Out[3]: usm_ndarray(1.69314718)

In [4]: dpt.logaddexp(x2, x1)
Out[4]: usm_ndarray(nan)

This should return nan in both cases

[ndgrigorian]

Another case:

In [4]: x1 = dpt.asarray(dpt.inf)

In [5]: x2 = dpt.asarray(dpt.inf)

In [6]: dpt.logaddexp(x1, x2)
Out[6]: usm_ndarray(nan)

Should be dpt.inf

[ndgrigorian]

This issue was resolved in #1324