Implement helpers for stages of the learning loop

[ariasanovsky]

To do:

• [x] identify data parallel operations to ergonomify
  • [x] init s_0, c_t, trees
  • [x] fill x_0, x_0_dev, pi_t_theta, etc
  • [x] roll out episode
  • [ ] set up epoch
  • [x] instead, we need to reset epoch
  • [x] and update model
  • [x] set up learning loop
  • [x] select next roots
• [x] don't forget to make into_unstable_sorted_nodes(self) into unstable_sorted_nodes(&self)
• [x] create ready-mades for:
  • [x] AzModel for dfdx and for candle
  • [x] adding Dirichlet noise
  • [x] updating model

Structs

• [x] states: s_0, s_t, c_t, x_t
• [x] predictions: pi_t, g_t
• [x] model: tensors + models
• [x] trees: trees, p_t, transitions, nodes, ends
• [x] go through the get_mut's and encapsulate those which may unnecessarily create invalid states

Cargo.toml features

• [ ] candle: impl AzModel
  • moved to #52
• [x] dfdx: impl AzModel
• [x] tensorboard: output logging
• [x] rayon: parallel helpers
• [x] rand: adding noise to distributions

Questions

• [x] should there be a learning_loop struct? should it include the logger?
• [ ] do we want a runtime hyperparameter adjustment with ratatui?
  • tabled

[ariasanovsky]

Breakdown in progress...

[ariasanovsky]

Color-coding to identify intermediate structs

[ariasanovsky]

Converging on the breakdown, thought I'd share a pic too! Just need to get tree