bug: to_cupy_conditional device selection

[mrava87]

The to_cupy_conditional method creates cupy array on the current device instead of the device of x.

Proposed solution in #578:

def to_cupy_conditional(x: npt.ArrayLike, y: npt.ArrayLike) -> NDArray:
    if deps.cupy_enabled:
        if cp.get_array_module(x) == cp and cp.get_array_module(y) == np:
            with cp.cuda.Device(x.device):
                y = cp.asarray(y)
    return y

[mrava87]

Indeed, after the suggested fix, this code:

cp.cuda.Device(device=0).use()

xcurrent = cp.ones(10)

with cp.cuda.Device(1):
    xother = cp.ones(10)

print(xcurrent.device, xother.device)

y = np.zeros(5)

# from current device
ycurrent = to_cupy_conditional(xcurrent, y)

# from other device
yother = to_cupy_conditional(xother, y)

print(ycurrent.device, yother.device)

gives:

<CUDA Device 0> <CUDA Device 1>
<CUDA Device 0> <CUDA Device 1>