add train@ resume mode, refactor enjoy mode

discrete sac benchmark update

full benchmark upload

Soft Actor-Critic

atari benchmark, API simplification

add out_layer_activation

rework eval mode; major refactoring

add retro_eval, fix boltzmann spec

L1 fitness norm, code and spec refactor, online eval

PyTorch 1.0, faster NN, Variable Scheduler, working Atari