Refine Tmove v3 option algorithm

[ye-luo]

As updated in the manual.

The old algorithm

v3 reuses the transition probability computed in non-local pseudopotentials, namely before accepting any T-moves.

The new one

v3 mostly reuses the transition probability computed during the evaluation of non-local pseudopotentials for the local energy, namely before accepting any T-moves, and only recomputes the transition probability of the electrons within the same pseudopotential region of any electrons touched by T-moves.

v1 and v3 is not numerically equivalent but I got negligible difference in both molecules and solids with the new v3.

v3 faster algorithm will be the topic of a paper currently in preparation.

Additional changes:
New tests based on NiO are included. Closes #630
Merge the replicated code for CPU and GPU used by v0.

[prckent]

Thanks for the updates to this new algorithm.