Fix noncolin

• Fixes a a bug in for the noncolinear magnetic case.
• thread parallelization of compute_rho ( the routine use to create a local rho_up and rho_dn in the non colinear case)

tests:

• Fe Spinorbit with BLYP functional

  • reference calculation with spin orbit qe-641-spinOrb-blyp
  • develop version + this fix oBlyp-fix

• Fe Spinorbit with PBE functional

  • reference calculation qe-641_spinOrbit
  • develop version + this fix oPBE-fix

• Fe Scalar Relativistic with PBE functional

  • reference calculation qe-641-PBE_SR
  • develop version + this fix oPBE-SR-fix
  • PH at X reference qe-641-ph_PBE_SR
  • PH at X develop version + this fix oPH_PBE_SR