Fix out-of-bound errors in HP

The problem in hp_soc_U_nc_nonmagn should be fixed: the array becp1 (=<vkb|psi>) is now allocated also in the non-US case, allowing adddvscf to work properly (i.e. returning immediately because okvan=.false.).

The problem in hp_soc_UV_paw_magn is more mysterious, because it actually doesn't involve the noncollinear implementation (it concerns the post processing operations, which depend only on the crystal cell). However, it could be due to an unlucky choice of the input parameters of the test, rather than a bug of the code. I adjusted the test, let's see if this fixes the problem.