llm-proxy(app): gemma 反向代理 + token 鉴权 + /chat web UI

新 service，ns `llm-proxy`，域 `llm.famzheng.me`。
- POST /v1/chat/completions — OpenAI 兼容透传到 mochi 同款 backend
  gateway (gemma-4-31b-it)；一期强制 stream=false，SSE 留二期
- 鉴权: `Authorization: token <PROXY_AUTH_TOKEN>` 或同款 Bearer；
  常时间比较防 timing；空 expected 一律拒
- GET /chat — 自带极简 HTML chat UI（token 走 localStorage，
  附 curl example details）；/ 跳转到 /chat
- Secrets `llm-proxy/proxy-credentials` 已 kubectl 手工创建：
  BACKEND_TOKEN (上游) + PROXY_AUTH_TOKEN (对外)
- 13 个 cargo test 覆盖 auth 多个 scheme / 边界 + body
  改写 (stream=false 强制注入)

apps/llm-proxy/src/main.rs

@@ -0,0 +1,147 @@
+//! llm.famzheng.me — gemma-4-31b-it 反向代理 + 简单 token 鉴权。
+//!
+//! - `GET  /`                       → `/chat` 跳转
+//! - `GET  /chat`                   → 静态 web UI
+//! - `POST /v1/chat/completions`    → OpenAI 兼容透传 (要 Authorization: token <PROXY_AUTH_TOKEN>)
+//! - `GET  /healthz`                → 不带 auth, 给 k8s probe
+
+mod proxy;
+
+use std::sync::Arc;
+
+use axum::{
+    extract::State,
+    http::{header, StatusCode},
+    middleware::{self, Next},
+    response::{Html, IntoResponse, Redirect, Response},
+    routing::{get, post},
+    Router,
+};
+use tower_http::trace::TraceLayer;
+
+#[tokio::main]
+async fn main() -> std::io::Result<()> {
+    cube_core::init_tracing();
+    let cfg = Arc::new(proxy::Config::from_env());
+    let port: u16 = std::env::var("PORT")
+        .ok()
+        .and_then(|s| s.parse().ok())
+        .unwrap_or(8080);
+
+    let chat_api = Router::new()
+        .route("/v1/chat/completions", post(proxy::handle))
+        .route_layer(middleware::from_fn_with_state(cfg.clone(), require_token))
+        .with_state(cfg);
+
+    let app = Router::new()
+        .route("/healthz", get(|| async { "ok" }))
+        .route("/", get(|| async { Redirect::permanent("/chat") }))
+        .route("/chat", get(chat_ui))
+        .merge(chat_api)
+        .layer(TraceLayer::new_for_http());
+
+    let addr = format!("0.0.0.0:{port}");
+    let listener = tokio::net::TcpListener::bind(&addr).await?;
+    tracing::info!(%addr, "llm-proxy listening");
+    axum::serve(listener, app).await
+}
+
+const CHAT_HTML: &str = include_str!("../web/chat.html");
+
+async fn chat_ui() -> Html<&'static str> {
+    Html(CHAT_HTML)
+}
+
+/// 验 `Authorization: token <PROXY_AUTH_TOKEN>`，错的直接 401。
+async fn require_token(
+    State(cfg): State<Arc<proxy::Config>>,
+    req: axum::extract::Request,
+    next: Next,
+) -> Response {
+    let header_val = req
+        .headers()
+        .get(header::AUTHORIZATION)
+        .and_then(|v| v.to_str().ok())
+        .map(str::trim);
+
+    match header_val {
+        Some(v) if check_token(v, &cfg.proxy_auth_token) => next.run(req).await,
+        _ => (
+            StatusCode::UNAUTHORIZED,
+            "缺少或不匹配 `Authorization: token <your-token>`",
+        )
+            .into_response(),
+    }
+}
+
+/// 接受 `token <T>` 或 `Bearer <T>`（OpenAI client 习惯发 Bearer，宽容点）。
+pub fn check_token(header_value: &str, expected: &str) -> bool {
+    if expected.is_empty() {
+        return false;
+    }
+    let trimmed = header_value.trim();
+    if let Some(rest) = trimmed.strip_prefix("token ") {
+        return constant_time_eq(rest.trim().as_bytes(), expected.as_bytes());
+    }
+    if let Some(rest) = trimmed.strip_prefix("Bearer ") {
+        return constant_time_eq(rest.trim().as_bytes(), expected.as_bytes());
+    }
+    false
+}
+
+/// 常时间比较，防 timing attack（虽然这场景影响小，做了不亏）。
+fn constant_time_eq(a: &[u8], b: &[u8]) -> bool {
+    if a.len() != b.len() {
+        return false;
+    }
+    let mut diff: u8 = 0;
+    for (x, y) in a.iter().zip(b.iter()) {
+        diff |= x ^ y;
+    }
+    diff == 0
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn check_token_accepts_token_scheme() {
+        assert!(check_token("token famzheng-llm-2026", "famzheng-llm-2026"));
+    }
+
+    #[test]
+    fn check_token_accepts_bearer_scheme() {
+        assert!(check_token("Bearer famzheng-llm-2026", "famzheng-llm-2026"));
+    }
+
+    #[test]
+    fn check_token_rejects_wrong_value() {
+        assert!(!check_token("token wrong", "famzheng-llm-2026"));
+    }
+
+    #[test]
+    fn check_token_rejects_unknown_scheme() {
+        assert!(!check_token("Basic famzheng-llm-2026", "famzheng-llm-2026"));
+        assert!(!check_token("famzheng-llm-2026", "famzheng-llm-2026"));
+    }
+
+    #[test]
+    fn check_token_rejects_empty_expected() {
+        // 防 misconfigured：空 expected 不应该让任何人通过
+        assert!(!check_token("token any", ""));
+        assert!(!check_token("Bearer ", ""));
+    }
+
+    #[test]
+    fn check_token_strips_extra_whitespace() {
+        assert!(check_token("  token   famzheng-llm-2026  ", "famzheng-llm-2026"));
+    }
+
+    #[test]
+    fn check_token_rejects_prefix_match() {
+        // 防止"famzheng-llm-2026-extra" 通过
+        assert!(!check_token("token famzheng-llm-2026-extra", "famzheng-llm-2026"));
+        assert!(!check_token("token famzheng-llm", "famzheng-llm-2026"));
+    }
+}