Refactors policy/q-functions, pushes old code to deprecated, adds new optimizations/modular dendritic layers

[abhi-iyer]

• Contains optimizations to AbsoluteMaxGatingDendriticLayer:

  • Used amax instead of max to speed up gradient step
  • Used **2 instead of abs() during eval mode

• All code in algorithms/, modules/, value_functions/, policies/, and dendrites/ now

• Added functionality for modular dendritic layers in RL policy & value functions (i.e. only modulate specific layers with dendrites)

• Added functionality for automatically changing input and context information into policy & value functions

• Move old code to deprecated folder (in nupic.embodied/ and projects/)

• Clean up parser, specifically NetworkArguments

• Resolve code ambiguities and unnecessary indirections