refactor: 训练循环改为两重迭代并统一参数命名

- 训练循环从三重(epoch→batched→batch)改为二重(epoch→batch)
- batch_size → batch_per_device, accumulation_steps → grad_accum_steps
- scheduler 移入 step block 对齐 optimizer 更新步
- GradientClippingCallback 改用 on_step_begin 避免零梯度裁剪
- 移除 _train_impl 误导性的 -> Checkpoint 标注
- total_steps 修除为向下取整并精简为一行
- warmup_steps 改为 warmup_ratio (默认0.05)

astrai/trainer/train_callback.py

@@ -79,8 +79,7 @@ class GradientClippingCallback(TrainCallback):
     def __init__(self, max_grad_norm: float):
         self.max_grad_norm = max_grad_norm
 
-    def on_step_end(self, context: TrainContext):
-        _ = context
+    def on_step_begin(self, context: TrainContext):
         clip_grad_norm_(context.model.parameters(), self.max_grad_norm)