cupy 13.5 vs pytorch

[gschramm]

import cupy as cp
import array_api_compat.torch as xp

a = xp.asarray(cp.ones(5)

throws an error.

Needed e.g. in the Gaussian filter and projectors when projecting torch tensors.
Might need to go via DLpack.

[gschramm]

use from_dplack() should work if we stick to array_api_compat