feat: 故障推送 / 公需课自动重学 / 插件解耦，发版 v0.0.8

体检修复：启动日志不再打印 AUTH_TOKEN、密钥可清空、推送失败不占整段冷却、目录名与文案统一成 plugins/。

src/study_tool/routes/llm.py

@@ -0,0 +1,29 @@
+"""大模型连通性测试接口。"""
+import time
+
+from study_tool.webapp import app
+
+from study_tool.config.settings import load_settings
+from study_tool.llm import llm_chat
+from study_tool.models import LlmTestIn
+
+
+@app.post("/api/llm/test")
+def api_llm_test(b: LlmTestIn):
+    cfg = dict(load_settings()["llm"])
+    if b.base_url.strip():
+        cfg["base_url"] = b.base_url.strip()
+    if b.model.strip():
+        cfg["model"] = b.model.strip()
+    if b.api_key.strip():
+        cfg["api_key"] = b.api_key.strip()
+    cfg["timeout"] = max(10, min(120, int(b.timeout or 60)))
+    if not cfg.get("api_key"):
+        return {"ok": False, "error": "没有 api_key"}
+    t0 = time.time()
+    try:
+        reply = llm_chat([{"role": "user", "content": "只回复两个字符：ok"}], cfg,
+                         max_tokens=8, json_mode=False)
+        return {"ok": True, "ms": int((time.time() - t0) * 1000), "reply": (reply or "")[:80]}
+    except Exception as e:
+        return {"ok": False, "error": str(e)[:300]}