Make sleap-train command able to accept more than one config for top-down

[eberrigan]

• The top-down pipeline in SLEAP requires two configs. Right now the process training command is written with the bottom-up pipeline in mind. We have the config path hardcoded except for the version for our MLOps pipeline use case.

  sleap-roots-training/sleap_roots_training/train.py

  Line 363 in 72b580c

  def process_training(project_name, entity_name, experiment_name, version, group, use_existing_model, sleap_train_command, tags=None, model_tags=None):

• Let's add a top-down option, as a boolean that's false by default. When true there will be a second config for the centered instance model.
• The centroid model naming convention for the training config is

  sleap-roots-training/sleap_roots_training/train.py

  Line 383 in 72b580c

  config_path = dir_path / f"initial_config_modified_v00{version}.json"

  .
• The centered instance model will follow a naming convention that is similar.
  • dir_path / f"centered_instance_initial_config_modified_v00{version}.json"
• Test the pipeline with top down models and see if anything else needs to be modified.
• We should add helper notebooks for top-down training.