semi-local lambda lifting in holyhammer translation + s-expression export for tree neural networks

missing tttRecord. signature in tttUnfold code

summing the gradients for the node before backpropagating

precomputation before starting big steps

faster ordering of term list

more parallelization during neural network training (does not seem to help in the average case)

new reinforcement learning loop

before removing external parallelism training failed experiment

removing trailing whitespaces + fix cache errors in term generation

aim model: more efficient testing of inner mappings properties

nn training with external process (wip)

optimizing matrix addition

nn: I/O

nn: Xavier's initialization

modifying some filenames

refactoring tactictoe/holyhammer (wip)