feat: update pl version supports

### What does this PR do? Update for compatibility with pytorch-lightning 2.0.3 as well as backward compatibility with previous version (>1.6.4) Changes: - add `on_validation_epoch_end()` and remove `validation_epoch_end()` in `LightningModule` - `set_seed()` in `train()` - add `TQDMProgressBar` in `callbacks` argument in `Trainer()`, and remove `get_progress_bar_dict()` in `LightningModule` - add `ckpt_path` argument in `Trainer.fit()`, and remove `resume_from_checkpoint` argument in `Trainer()` Minor changes: - remove `self.model.norm` in `SwinEncoder` to free the unused parameters - add `self.validation_step_outputs` variable in `LightningModule` for `on_validation_epoch_end()` - change the `gpus` argument to `devices` argument in `Trainer()`