Rework configuration classes

[jlamypoirier]

A first pass at reworking the configuration classes to avoid redundancies, inconsistancies, confusing names, etc.
Also chopped off much of the obsolete RunConfig.

RENAME,DEFAULT training.validation_interval = 1000 -> training.validation.interval = None
RENAME,DEFAULT training.validation_iters = 0 -> training.validation.iterations = None
ADD training.validation.offset = 0
OTHER: Must set both interval > 0 and iterations > 0 to enable validation.

RENAME,DEFAULT run.log_interval = 100 -> training.logs.interval = None
RENAME,DEFAULT run.log_offset = 1 -> training.logs.offset = 0

RENAME run.checkpoint_interval -> training.checkpoint.interval
RENAME run.checkpoint_offset -> training.checkpoint.offset
RENAME,DEFAULT run.max_checkpoints = None -> training.checkpoint.keep = 5

RENAME run.export_interval -> training.export.interval
ADD training.export.offset = 0
OTHER: Export must be sub-interval of checkpoint

RENAME, FORMAT run.export_callback_script:str -> training.export.callback.script:list[str]|None
RENAME, FORMAT run.export_callback_env:str -> training.export.callback.environment:dict

RENAME run.stop_interval -> training.shutdown.interval
RENAME run.stop_offset -> training.shutdown.offset
OTHER: Shutdown must be sub-interval of checkpoint

RENAME run.wandb_group_name -> training.wandb.group_name
RENAME run.wandb_project_name -> training.wandb.project_name
RENAME run.wandb_entity_name -> training.wandb.entity_name
RENAME run.wandb_status_interval -> training.wandb.alert.interval
ADD training.wandb.alert.offset = 0
RENAME run.wandb_post_alerts -> training.wandb.alert.status_updates
OTHER: Alerts must be sub-interval of logs
OTHER: Extract wandb into separate class

RENAME run.save_tensor_logs -> training.tensor_logs.save
RENAME run.show_tensor_logs -> training.tensor_logs.show
RENAME run.tensor_logs_show_elements -> training.tensor_logs.max_elements

RENAME pretrained.pretrained_checkpoint_path -> pretrained.path
RENAME pretrained.pretrained_checkpoint_type -> pretrained.format
RENAME pretrained.imported_model_type -> pretrained.imported_type
REMOVE pretrained.use_pretrained_config
RENAME pretrained.ignore_pretrained_config -> pretrained.override_architecture
RENAME pretrained.load_pretrained_weights -> pretrained.load_weights
RENAME pretrained.load_pretrained_optimizer -> pretrained.load_optimizer

RENAME data.tokenizer.tokenizer_type -> data.tokenizer.format
RENAME data.tokenizer.tokenizer_file -> data.tokenizer.path

RENAME data.fim.fim_[...] -> data.fim.[...]

REMOVE data.dataset_type
RENAME data.dataset_source -> data.format
RENAME data.data_path -> data.path

RENAME profile.profile_column_width -> profile.table_width
RENAME profile.profile_[...] -> profile.[...]

RENAME optimizer.lr_schedule.lr -> optimizer.learning_rate.base
RENAME optimizer.lr_schedule.lr_decay_style -> optimizer.learning_rate.decay_style
RENAME optimizer.lr_schedule.lr_decay_iters -> optimizer.learning_rate.decay_iterations
RENAME optimizer.lr_schedule.lr_decay_power -> optimizer.learning_rate.decay_power
RENAME optimizer.lr_schedule.lr_warmup_iters -> optimizer.learning_rate.warmup_iterations
RENAME optimizer.lr_schedule.min_lr -> optimizer.learning_rate.minimum
RENAME optimizer.lr_schedule.lr_schedule -> optimizer.learning_rate.schedule

RENAME optimizer.adam_beta1 -> optimizer.beta_1
RENAME optimizer.adam_beta2 -> optimizer.beta_2
RENAME optimizer.adam_eps -> optimizer.epsilon

RENAME optimizer.clip_grad -> optimizer.gradient_norm_clipping
RENAME optimizer.loss_scale -> optimizer.gradient_scaler.constant
RENAME optimizer.initial_loss_scale -> optimizer.gradient_scaler.initial
RENAME optimizer.min_loss_scale -> optimizer.gradient_scaler.minimum
RENAME optimizer.loss_scale_window -> optimizer.gradient_scaler.window
RENAME optimizer.hysteresis -> optimizer.gradient_scaler.hysteresis
RENAME optimizer.default_lr_scale -> optimizer.default_learning_rate_scale
REMOVE optimizer.lr_schedule_offset

!!!RENAME model.base_model.transformer.normalization.normalization_type -> model.base_model.transformer.normalization.type
!!!RENAME model.base_model.transformer.normalization.layer_norm_eps -> model.base_model.transformer.normalization.epsilon
!!!RENAME model.base_model.transformer.normalization.zero_centered_normalization -> model.base_model.transformer.normalization.zero_centered
RENAME model.base_model.transformer.normalization.normalization_implementation -> model.base_model.transformer.normalization.implementation
RENAME model.base_model.transformer.normalization.layer_norm_init_range -> model.base_model.transformer.normalization.initialization_range