refactor: 提取 HandlerAdapterBase 基类，新增 api_family/endpoint_kind 结构化维度

- 从 ChatAdapterBase 和 CliAdapterBase 提取公共逻辑到 HandlerAdapterBase，
  消除头部处理、异常处理、计费策略等重复代码
- Usage 记录链路全程透传 api_family / endpoint_kind / provider_api_family /
  provider_endpoint_kind 四个结构化字段，替代从 api_format 字符串解析
- 删除冗余的 ClaudeCliNormalizer、GeminiCliNormalizer、ClaudeCliResponseParser、
  GeminiCliResponseParser，改用 data_format_id 回退机制自动复用 Chat 版本
- build_endpoint_url 签名统一扩展 request_data / model_name 参数
- 新增 Alembic 迁移，含历史数据回填

src/core/api_format/conversion/normalizers/claude_cli.py

@@ -1,17 +0,0 @@
-"""
-Claude CLI Normalizer
-
-CLAUDE_CLI 的请求/响应 body 与 CLAUDE 一致（Anthropic Messages API），差异主要在认证头。
-因此这里复用 ClaudeNormalizer 的转换逻辑，仅更换 FORMAT_ID。
-
-如需 CLI 特殊处理，可覆盖 request_from_internal / request_to_internal 等方法。
-"""
-
-from src.core.api_format.conversion.normalizers.claude import ClaudeNormalizer
-
-
-class ClaudeCliNormalizer(ClaudeNormalizer):
-    FORMAT_ID = "claude:cli"
-
-
-__all__ = ["ClaudeCliNormalizer"]

src/core/api_format/conversion/normalizers/gemini_cli.py

@@ -1,17 +0,0 @@
-"""
-Gemini CLI Normalizer
-
-GEMINI_CLI 的请求/响应 body 与 GEMINI 一致（Google Gemini API），差异主要在鉴权/UA 等请求层。
-因此这里复用 GeminiNormalizer 的转换逻辑，仅更换 FORMAT_ID。
-
-如需 CLI 特殊处理，可覆盖 request_from_internal / request_to_internal 等方法。
-"""
-
-from src.core.api_format.conversion.normalizers.gemini import GeminiNormalizer
-
-
-class GeminiCliNormalizer(GeminiNormalizer):
-    FORMAT_ID = "gemini:cli"
-
-
-__all__ = ["GeminiCliNormalizer"]

src/core/api_format/conversion/registry.py

@@ -54,7 +54,22 @@ class FormatConversionRegistry:
         logger.info(f"[FormatConversionRegistry] 注册 normalizer: {normalizer.FORMAT_ID}")
 
     def get_normalizer(self, format_id: str) -> FormatNormalizer | None:
-        return self._normalizers.get(str(format_id).upper())
+        key = str(format_id).upper()
+        # 1. 精确匹配
+        normalizer = self._normalizers.get(key)
+        if normalizer is not None:
+            return normalizer
+        # 2. data_format_id 回退：如 "claude:cli" (dfid="claude") -> ClaudeNormalizer (dfid="claude")
+        from src.core.api_format.metadata import get_data_format_id_for_endpoint
+
+        target_dfid = get_data_format_id_for_endpoint(format_id)
+        if not target_dfid:
+            return None
+        for reg_key, reg_normalizer in self._normalizers.items():
+            reg_dfid = get_data_format_id_for_endpoint(reg_key)
+            if reg_dfid == target_dfid:
+                return reg_normalizer
+        return None
 
     def _require_normalizer(self, format_id: str) -> FormatNormalizer:
         normalizer = self.get_normalizer(format_id)

src/core/stream_types.py

@@ -219,6 +219,16 @@ def get_parser_for_format(format_id: str) -> ResponseParser:
         )
 
     normalized = normalize_signature_key(format_id)
-    if normalized not in _PARSER_REGISTRY:
-        raise KeyError(f"Unknown format: {normalized}")
-    return _PARSER_REGISTRY[normalized]()
+    # 1. 精确匹配
+    if normalized in _PARSER_REGISTRY:
+        return _PARSER_REGISTRY[normalized]()
+    # 2. data_format_id 回退：如 "claude:cli" (dfid="claude") -> ClaudeResponseParser (dfid="claude")
+    from src.core.api_format.metadata import get_data_format_id_for_endpoint
+
+    target_dfid = get_data_format_id_for_endpoint(normalized)
+    if target_dfid:
+        for reg_key, parser_cls in _PARSER_REGISTRY.items():
+            reg_dfid = get_data_format_id_for_endpoint(reg_key)
+            if reg_dfid == target_dfid:
+                return parser_cls()
+    raise KeyError(f"Unknown format: {normalized}")