Proposal for codebase reorganization

[fmassa]

The current structure of the repo is as follows

We are starting to have a number of files in the package root, and it might make it harder to find what we want.

I'm proposing to restructure the repo as follows:

autoparallel/
  _testing/models/

  graph_passes/
    activation_checkpointing.py
    async_tp/
    autobucketing_inductor/
    graph_clustering.py
    graph_utils.py

  shardings/
    propagation_rules.py
    utils.py -> rename to placement_options.py 
    dtensor_util/utils.py -> rename to something else and delete dtensor_util

  cost_models/
    compute_estimation.py
    collective_runtime_estimation.py

  ordered_sharding.py
  api.py
  apply_sharding.py
  cast_parametrization.py
  init_weights.py
  optimize_sharding.py
  collectives.py

[ezyang]

sgtm