Add TensorBoard support for training metrics logging

[iqiancheng]

Summary

Add TensorBoard support for training metrics logging alongside existing Wandb support. #187

Changes

• Add unified Logger class (veomni/utils/metric_logger.py): Support both Wandb and TensorBoard logging with a single interface
• Add TensorBoard configuration options (veomni/utils/arguments.py):
  • use_tensorboard: Enable/disable TensorBoard logging
  • tensorboard_dir: Directory to save TensorBoard logs
• Update training scripts: Refactor all training scripts (train_flux.py, train_qwen_vl.py, train_qwen2_vl.py, train_omni_model.py, train_wan.py, train_torch.py) to use the unified Logger
• Add documentation (docs/tutorials/tensorboard_and_wandb.md) link: Guide on how to set up and use both TensorBoard and Wandb

Usage

Users can now enable TensorBoard logging by adding to their config:

train:
  use_tensorboard: true
  tensorboard_dir: "./output/tensorboard"

Both TensorBoard and Wandb can be enabled simultaneously for local and cloud-based experiment tracking.

Testing

• Verified TensorBoard logging works correctly
• Verified Wandb logging remains functional
• Verified both can be enabled simultaneously

[Luosuu]

Hi @iqiancheng , thanks for your contribution. This looks big. Can you explain why you would like to use tensorboard?