Spin-polarized ADF calculations should not use a DFTB initial hessian

[YHordijk]

Currently when you specify that a calculation should be an optimization or transition state search the AMSJob class will automatically add the initialhessian key to the settings. For ADF calculations this uses DFTB, which is usually fine, but in the case of spin-polarized ADF calculation this is in principle not a good idea. This is because DFTB does not support the use of spin-polarization and may then give a completely wrong hessian.
For now, it might be good to disable this initialhessian when spin-polarization is used. In the future we can then maybe add a step where it calculates the hessian with a very cheap DFT calculation instead.