Refactor server.py: add base64 support, separate from training

- POST /solve accepts JSON with base64 encoded image
- POST /solve/upload keeps multipart file upload compatibility
- Server only depends on inference code (onnxruntime), no torch
- Catch invalid image errors with proper 400 response
- Update CLAUDE.md with new API documentation

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

CLAUDE.md

@@ -440,11 +440,19 @@ uv run python cli.py serve --port 8080
 
 ## HTTP 服务 (server.py，可选)
 
+纯推理服务，不依赖 torch / 训练代码，仅需 onnxruntime + FastAPI。
+
 ```python
-# FastAPI 服务，提供 REST API
-# POST /solve  - 上传图片，返回识别结果
-# 请求: multipart/form-data，字段名 image
-# 响应: {"type": "normal", "result": "A3B8", "confidence": 0.95, "time_ms": 45}
+# POST /solve         - JSON base64 图片识别
+#   请求: {"image": "<base64>", "type": "normal"}   (type 可选)
+#   响应: {"type": "normal", "result": "A3B8", "raw": "A3B8", "time_ms": 12.3}
+#
+# POST /solve/upload  - multipart 文件上传识别
+#   请求: multipart/form-data, 字段名 image, 可选 query param type
+#   响应: 同上
+#
+# GET  /health        - 健康检查
+#   响应: {"status": "ok", "models_loaded": true}
 ```
 
 ## 关键约束和注意事项