docs: 更新文档与代码同步（Executor/训练循环/参数）

- architecture.md: TrainConfig 移除旧 parallel_wrapper/state_dict_fn
- architecture.md: 新增 ExecutorFactory/BaseExecutor/DDPExecutor 等类图
- architecture.md: MLA 新增 use_qk_norm/q_norm/k_norm
- architecture.md: 新增 protocols 命名空间
- training.md: 修复训练循环 hook 名和 scheduler.step 位置
- training.md: 替换 parallel_wrapper 为 parallel_mode/executor.prepare
- training.md: 修复默认回调顺序和 Callback 生命周期表
- params.md: 新增 --parallel_mode 和 --start_method

assets/docs/params.md

@@ -53,7 +53,9 @@
 | Parameter | Description | Default |
 |-----------|-------------|---------|
 | `--nprocs` | Number of GPUs / processes | 1 |
+| `--parallel_mode` | Parallel strategy (`none` or `ddp`) | none |
 | `--device_type` | Device type | cuda |
+| `--start_method` | Multiprocessing start method (`spawn`, `fork`, `forkserver`) | spawn |
 
 ### Strategy-specific
 
@@ -94,4 +96,4 @@ nohup python scripts/tools/train.py \
 
 ---
 
-> Document Update Time: 2026-05-17
+> Document Update Time: 2026-05-24