Inexact Krylov methods and adaptive response

[mfherbst]

Will superseed #1027 and #737:

• Preconditioned GMRES implementation ok ?
• Factor of 2 next to fn in Bonan's implementation ?
• Balanced strategy: For the orbital term lower bound we should not use the total Nocc
• Think about what happens if we use q ≠ 0. Are things ok ?
• Testing: A more challenging test case (e.g. aluminium ?)
• Use default SCF mixing as preconditioner
• Cleanup code a bit more
• Nicer print formatting
• Example