Possible Performance Optimization

[shaomeng]

1. It appears that the NewlySig enum is unnecessary.
2. Consider refine all data points all together, and use a bitmask to save the output of whether individual points are significant or not. This is equivalent to the current approach, but I suspect that it's gonna be faster.
   • To clarify, in the current code, a data point is refined first in the m_process_P or m_process_P_lite routine, and then repeatedly in the m_refinement_pass routine.
   • On the one hand, it should be helpful to really refine the data buffer in only one place, and simply use a bitmask to keep necessary information in other places.
   • On the other hand, a significant data point is already located, so refining it wouldn't cost another random access. The benefit of this approach really needs to be evaluated.