docs: 同步 architecture/inference/training 文档至实际代码，CLI 补充 fsdp 选项

- 修正 ProtocolHandler 架构：concrete + ResponseBuilder(ABC) 策略模式
- 修正训练循环 scheduler.step() 在 sync_gradients 块内
- 修正组合/聚合关系：注入组件改为 o--，删除不持有引用的关联
- --parallel_mode CLI choices 加入 fsdp
- nprocs > 1 且 parallel_mode=none 时 raise error

scripts/tools/train.py

@@ -147,8 +147,8 @@ def parse_args() -> argparse.Namespace:
         "--parallel_mode",
         type=str,
         default="none",
-        choices=["none", "ddp"],
-        help="Parallel training strategy.",
+        choices=["none", "ddp", "fsdp"],
+        help="Parallel training strategy (none, ddp, fsdp).",
     )
     parser.add_argument(
         "--device_type", type=str, default="cuda", help="Device type to use."
@@ -228,6 +228,8 @@ def train(
 ):
     assert train_type in ["seq", "sft", "dpo", "grpo"]
     assert os.path.exists(param_path)
+    if nprocs > 1 and parallel_mode == "none":
+        raise ValueError("--nprocs > 1 requires --parallel_mode to be 'ddp' or 'fsdp'")
 
     # Load config
     config_path = os.path.join(param_path, "config.json")