feat(web): settings with 9 sidebar sections and Agent upstream management

- /console/settings#<section>[/<anchor>]; legacy ?tab= redirected in clientLoader
- SETTINGS_SECTIONS in contracts with field-ownership test
- Agent section: upstream list, create wizard (provider -> credentials/model -> login/test/default),
  edit drawer with OAuth browser/device login, test, set default, delete
- sites section with site registration, snapshot/rate-limit drawer and account credentials
- path mappings moved into downloader / media-server drawers; metadata and account sections

Co-Authored-By: Claude Opus 5.5 <[email protected]>

apps/web/app/components/settings/pi-test.tsx

@@ -0,0 +1,72 @@
+import type { PiUpstream, PiUpstreamTestResult } from "@umt/contracts";
+import { LoaderCircle, PlugZap, Send } from "lucide-react";
+import { useState } from "react";
+import { toast } from "sonner";
+import { useTestPiUpstream } from "~/api/pi.ts";
+import { StatusBadge } from "~/components/console/status-badge.tsx";
+import { Button } from "~/components/ui/button.tsx";
+import { errorMessage } from "~/lib/api.ts";
+
+// 上游测试:默认只验凭据与模型(不产生费用);"发一次最小补全"真实调用一次模型
+
+const ERROR_KIND: Record<string, string> = {
+  auth: "凭据被拒",
+  oauth: "登录失效",
+  model: "模型不可用",
+  network: "网络错误",
+  config: "配置错误",
+  other: "失败",
+};
+
+export function PiTestPanel({ upstream, model }: { upstream: PiUpstream; model?: string | null }) {
+  const test = useTestPiUpstream();
+  const [result, setResult] = useState<PiUpstreamTestResult | null>(null);
+  const [running, setRunning] = useState<"auth" | "completion" | null>(null);
+
+  const run = async (mode: "auth" | "completion") => {
+    setRunning(mode);
+    try {
+      const r = await test.mutateAsync({ id: upstream.id, body: { mode, ...(model ? { model } : {}) } });
+      setResult(r);
+      if (r.ok) toast.success(mode === "auth" ? "凭据可用" : "补全成功");
+      else toast.error(r.message ?? "测试失败");
+    } catch (err) {
+      setResult(null);
+      toast.error(errorMessage(err, "测试失败"));
+    } finally {
+      setRunning(null);
+    }
+  };
+
+  return (
+    <div className="grid gap-2">
+      <div className="flex flex-wrap gap-2">
+        <Button type="button" size="sm" variant="outline" disabled={!!running} onClick={() => run("auth")}>
+          {running === "auth" ? <LoaderCircle className="animate-spin" /> : <PlugZap />}
+          测试
+        </Button>
+        <Button type="button" size="sm" variant="ghost" disabled={!!running} onClick={() => run("completion")}>
+          {running === "completion" ? <LoaderCircle className="animate-spin" /> : <Send />}
+          发一次最小补全
+        </Button>
+      </div>
+      <p className="text-meta text-muted-foreground">
+        「测试」只检查凭据和模型,不产生费用;最小补全会真实调用一次模型(最多 16 个 token)。
+      </p>
+      {result ? (
+        <div className="flex flex-wrap items-center gap-2 text-caption">
+          <StatusBadge tone={result.ok ? "success" : "destructive"}>
+            {result.ok ? "通过" : ERROR_KIND[result.errorKind ?? "other"]}
+          </StatusBadge>
+          <span className="text-muted-foreground">
+            {result.latencyMs} ms{result.model ? ` · ${result.model}` : ""}
+            {result.source ? ` · ${result.source}` : ""}
+          </span>
+          {!result.ok && result.message ? (
+            <span className="w-full break-words text-destructive">{result.message}</span>
+          ) : null}
+        </div>
+      ) : null}
+    </div>
+  );
+}