add qlearning writeup

even dumber - constant identity. Still doesn't work.

seemingly fixed but doesn't converge - I understand the code now

backed myself into a corner: dqn for discrete, ppo (or policy gradient stuff in general) for continuous

dummy env, doesn't work

cleanup

holy moly lunarlander training runs (but doesn't work)

Lint + add and modify the rest of the demo code

add brainstorming / spec to readme

modify FCN and memory for my purposes

Starting point: LunarLander, PyTorch tutorial