refactor: Config序列化统一BaseConfig基类

- 新增astrai/config/base.py，提供to_dict/from_dict基类
- 统一命名：load/save → from_file/to_file
- Checkpoint.meta合并训练配置到meta.json
- sys.stderr.warn → warnings.warn
- from_file改为classmethod

assets/docs/architecture.md

@@ -5,10 +5,15 @@
 ```mermaid
 classDiagram
     namespace config {
+        class BaseConfig {
+            +to_dict() Dict
+            +from_dict(d) Self
+        }
+
         class BaseModelConfig {
             +Optional[str] model_type
-            +load(config_path) Self
-            +save(config_path)
+            +from_file(config_path) Self
+            +to_file(config_path)
         }
 
         class ModelConfig {
@@ -147,6 +152,7 @@ classDiagram
             +int epoch
             +int iteration
             +dict extra
+            +dict meta
             +save(save_dir)
             +load(save_dir) Checkpoint
         }
@@ -750,6 +756,8 @@ classDiagram
     ParallelModel <|-- RowParallelLinear
     ParallelModel <|-- ColumnParallelLinear
     AutoModel <|-- Transformer
+    BaseConfig <|-- BaseModelConfig
+    BaseConfig <|-- TrainConfig
     BaseModelConfig <|-- ModelConfig
     BaseFactory <|-- AutoModel
     BaseFactory <|-- AttnFactory
@@ -838,7 +846,7 @@ classDiagram
 
 | Module | Components | Description |
 |--------|------------|-------------|
-| **astrai.config** | ModelConfig, TrainConfig | Configuration management |
+| **astrai.config** | BaseConfig, BaseModelConfig, ModelConfig, TrainConfig | Configuration management (to_dict/from_dict, to_file/from_file) |
 | **astrai.dataset** | BaseDataset–GRPODataset, BaseStorage–JSONStorage, BaseSegmentFetcher, MultiSegmentFetcher, ResumableDistributedSampler, DatasetFactory | Dataset loading and management |
 | **astrai.serialization** | Checkpoint | Model serialization |
 | **astrai.model** | AutoModel, Transformer, DecoderBlock, GQA, MLA, MLP, DeepSeekMoE, AttnFactory, FFNFactory, RMSNorm, Linear, RotaryEmbedding, Embedding | Neural network model |

assets/docs/training.md

@@ -157,12 +157,13 @@ Created by `SchedulerFactory.create(optimizer, schedule_type, **kwargs)`.
 ## Checkpoint
 
 ```
-Checkpoint(state_dict, epoch, iteration, extra)
-  ├── save(save_dir)    rank-0 only: meta.json + state_dict.safetensors + optional extra.pt
+Checkpoint(state_dict, epoch, iteration, extra, meta)
+  ├── save(save_dir)    rank-0 only: meta.json (includes training config) + state_dict.safetensors + optional extra.pt
   └── load(save_dir)    broadcasts metadata from rank-0
 ```
 
-Optimizer/scheduler state persisted by default via `Checkpoint.extra`.
+Optimizer/scheduler state persisted by default via `Checkpoint.extra`.  
+Training config (`TrainConfig.to_dict()`) saved into `meta.json` during training via `CheckpointCallback`.
 
 ## TrainContextBuilder (Builder Pattern)