feat: HttpLLMClient（httpx + 重试退避/超时/鉴权）

src/genesis/inference/client.py

@@ -0,0 +1,96 @@
+from __future__ import annotations
+
+import time
+from typing import Protocol, Sequence
+
+import httpx
+
+from .exceptions import LLMNetworkError, LLMNotConfiguredError, LLMTimeoutError
+from .types import ChatMessage, TokenUsage
+
+
+class LLMClient(Protocol):
+    """LLM 调用适配器（可注入替换为 Fake）。"""
+
+    def chat(
+        self,
+        *,
+        model: str,
+        messages: list[ChatMessage],
+        temperature: float,
+        max_tokens: int,
+    ) -> tuple[str, TokenUsage]: ...
+
+
+class HttpLLMClient:
+    """OpenAI Chat Completions 兼容的 httpx 实现；支持重试（指数退避）。"""
+
+    def __init__(
+        self,
+        *,
+        base_url: str,
+        api_key: str,
+        timeout_sec: float = 60.0,
+        retry_backoff: Sequence[float] = (1.0, 3.0, 7.0),
+        transport: httpx.BaseTransport | None = None,
+    ) -> None:
+        if not api_key:
+            raise LLMNotConfiguredError("LLM API key 未配置（DEEPSEEK_API_KEY / LLM_BASE_URL）")
+        self._base_url = base_url.rstrip("/")
+        self._api_key = api_key
+        self._timeout_sec = timeout_sec
+        self._retry_backoff = retry_backoff
+        self._client = httpx.Client(timeout=timeout_sec, transport=transport)
+
+    def chat(
+        self,
+        *,
+        model: str,
+        messages: list[ChatMessage],
+        temperature: float,
+        max_tokens: int,
+    ) -> tuple[str, TokenUsage]:
+        url = f"{self._base_url}/v1/chat/completions"
+        payload = {
+            "model": model,
+            "messages": [{"role": m.role, "content": m.content} for m in messages],
+            "temperature": temperature,
+            "max_tokens": max_tokens,
+        }
+        headers = {
+            "Authorization": f"Bearer {self._api_key}",
+            "Content-Type": "application/json",
+        }
+
+        attempts = 1 + len(self._retry_backoff)
+        last_error: Exception | None = None
+        for attempt in range(attempts):
+            if attempt > 0:
+                time.sleep(self._retry_backoff[attempt - 1])
+            try:
+                resp = self._client.post(url, json=payload, headers=headers)
+            except httpx.TimeoutException as exc:
+                last_error = exc
+                continue
+            except httpx.HTTPError as exc:
+                last_error = exc
+                continue
+
+            if resp.status_code >= 500:
+                last_error = LLMNetworkError(f"LLM 5xx: {resp.status_code}")
+                continue
+            if resp.status_code >= 400:
+                raise LLMNetworkError(f"LLM HTTP {resp.status_code}: {resp.text[:200]}")
+
+            data = resp.json()
+            content = data["choices"][0]["message"]["content"]
+            usage_raw = data.get("usage", {})
+            usage = TokenUsage(
+                input_tokens=usage_raw.get("prompt_tokens", 0),
+                output_tokens=usage_raw.get("completion_tokens", 0),
+            )
+            return content, usage
+
+        if isinstance(last_error, httpx.TimeoutException):
+            raise LLMTimeoutError(f"LLM 超时（{self._timeout_sec}s）") from last_error
+        raise LLMNetworkError(f"LLM 调用失败（重试耗尽）: {last_error}") from last_error