experiments 3-17

feat(config): update training configuration and dataset processing parameters

refactor(trainer): update dataloader typing and remove slow batch logging

feat(trainer): add in-memory batch loader with config updates

feat(dataset): add strided index support for efficient batch fetching

feat: add slow batch logging and timing measurements to DDP trainer

refactor(trainer): extract pin_memory logic into dedicated method with tests

feat(trainer): optimize DataLoader pin_memory for in-memory datasets with zero workers

feat(trainer): add CPU thread configuration for DDP training optimization

feat(config): increase batch size and update training data path

feat(runner): add CPU trainer selection and validation split handling

feat(trainer): add penalty support and autocast training for DDP models

feat: replace Trainer with TrainerDDP for distributed training support

feat: adjust dataset size estimation and model generation parameters

initial

minor fixes, tree cleanup

1-step output

minor output window fixes