refactor: 重构训练后端为 Executor 模式

- backend.py → executor.py，BaseTrainingBackend → BaseExecutor
- 新增 NoneExecutor（单卡）和 DDPExecutor（DDP，world_size=1 自动降级）
- 新增 GradientState 分离梯度同步状态，AccumOptimizer/AccumScheduler 包裹拦截
- 新增 astrai/protocols.py：OptimizerProtocol/SchedulerProtocol 结构子类型
- TrainContext.backend → executor，TrainConfig 移除 parallel_wrapper/state_dict_fn，新增 parallel_mode/executor_kwargs
- 训练循环用 accumulate() 包裹，on_optimizer_step 命名约定=gate
- scripts/tools/train.py 移除 ddp_wrap/prepare_checkpoint，新增 --parallel_mode

astrai/parallel/__init__.py

@@ -1,8 +1,10 @@
-from astrai.parallel.backend import (
+from astrai.parallel.executor import (
     AccumOptimizer,
     AccumScheduler,
-    BackendFactory,
-    BaseTrainingBackend,
+    BaseExecutor,
+    ExecutorFactory,
+    GradientState,
+    NoneExecutor,
 )
 from astrai.parallel.module import ColumnParallelLinear, RowParallelLinear
 from astrai.parallel.setup import (
@@ -23,8 +25,10 @@ __all__ = [
     "spawn_parallel_fn",
     "RowParallelLinear",
     "ColumnParallelLinear",
-    "BackendFactory",
-    "BaseTrainingBackend",
+    "ExecutorFactory",
+    "BaseExecutor",
+    "GradientState",
     "AccumOptimizer",
     "AccumScheduler",
+    "NoneExecutor",
 ]