This test for:
*GaAs-soc
*LCAO
*kpoints 2*2*2
*sg15 pseudopotential
*smearing_method gauss
*ks_solver genelpa
*mixing_type broyden-kerker
*mixing_beta 0.4

Compared with 240*SO, I added parameter soc_lambda = 0 to this test case,
which means I am performing the calculation in a nspin = 4 manner
but with soc strength = 0
Therefore, the result should be consistent with the calculation by turning off
soc (i.e., set lspinorb to be 0 in the INPUT file)
This is not the case for the old implementation of soc nonlocal PP (build_Nonlocal_mu)
but the new implementation (build_Nonlocal_mu_new) fixed it
