Dealing with CPU/GPU in QPFunctionFn's backward

[oumayb]

When the arguments of the QPFunctionFn's forward are on GPU, rhs[:dim] = dl_dzhat[I], rhs[dim : dim + neq] = dl_dlams[I] and rhs[dim + neq :] = dl_dnus[I] fail because rhs is a numpy array on CPU, and dl_dzhat , dl_dlams and dl_dnus are torch.tensors located on GPU.
I'm working on a PR to fix this.

proxsuite/bindings/python/proxsuite/torch/qplayer.py

Line 196 in 15cde91

rhs[:dim] = dl_dzhat[i]

PR: #297