refactor(proxy): preserve original header wire order and add non-streaming body timeout

- Rewrite build_raw_request to emit headers in original
  client-sent sequence instead of hash-map order
- Remove unused OriginalHeaderCases::get_all method
- Add body_timeout to read_decoded_body to prevent
  requests hanging when upstream stalls after headers

src-tauri/src/proxy/handlers.rs

@@ -226,7 +226,14 @@ async fn handle_claude_transform(
     }
 
     // 非流式响应转换 (OpenAI/Responses → Anthropic)
-    let (mut response_headers, _status, body_bytes) = read_decoded_body(response, ctx.tag).await?;
+    let body_timeout =
+        if ctx.app_config.auto_failover_enabled && ctx.app_config.non_streaming_timeout > 0 {
+            std::time::Duration::from_secs(ctx.app_config.non_streaming_timeout as u64)
+        } else {
+            std::time::Duration::ZERO
+        };
+    let (mut response_headers, _status, body_bytes) =
+        read_decoded_body(response, ctx.tag, body_timeout).await?;
 
     let body_str = String::from_utf8_lossy(&body_bytes);
 

src-tauri/src/proxy/hyper_client.rs

@@ -48,16 +48,6 @@ impl OriginalHeaderCases {
 
         Self { cases }
     }
-
-    /// Look up the original casing for a header name (case-insensitive).
-    /// Returns an iterator over all original casings for that name.
-    pub fn get_all<'a>(&'a self, name: &'a str) -> impl Iterator<Item = &'a [u8]> + 'a {
-        let lower = name.to_ascii_lowercase();
-        self.cases
-            .iter()
-            .filter(move |(k, _)| *k == lower)
-            .map(|(_, v)| v.as_slice())
-    }
 }
 
 type HyperClient = Client<
@@ -535,20 +525,53 @@ fn build_raw_request(
     raw.extend_from_slice(path_and_query.as_bytes());
     raw.extend_from_slice(b" HTTP/1.1\r\n");
 
-    // Headers with original casing
+    // Headers with original casing, emitted in original wire order.
-    for name in headers.keys() {
+    //
-        let name_str = name.as_str();
+    // Strategy:
-        let mut case_iter = original_cases.get_all(name_str);
+    // 1. Walk `original_cases.cases` in order — this preserves the exact
-        for value in headers.get_all(name) {
+    //    header sequence the client sent.  For each entry, emit the stored
-            if let Some(orig_name_bytes) = case_iter.next() {
+    //    original-casing name plus the current value from `headers` (the
+    //    proxy may have rewritten the value, e.g. Authorization).
+    //    Repeated headers with the same name are handled by tracking a
+    //    per-name value cursor so we step through `get_all()` in order.
+    // 2. After the original headers, append any headers that exist in
+    //    `headers` but were not present in the original request (i.e. added
+    //    by the proxy).  These are emitted in lowercase.
+    //
+    // This replaces the old `for name in headers.keys()` loop which iterated
+    // in hash-map order, destroying the original header sequence.
+    let mut emitted: std::collections::HashSet<String> =
+        std::collections::HashSet::with_capacity(original_cases.cases.len());
+    // Per-name cursor: how many values we have already emitted for each name.
+    let mut value_cursor: std::collections::HashMap<String, usize> =
+        std::collections::HashMap::with_capacity(original_cases.cases.len());
+
+    for (lower_name, orig_name_bytes) in &original_cases.cases {
+        if let Ok(header_name) = http::header::HeaderName::from_bytes(lower_name.as_bytes()) {
+            let all_values: Vec<_> = headers.get_all(&header_name).iter().collect();
+            let cursor = value_cursor.entry(lower_name.clone()).or_insert(0);
+            if let Some(value) = all_values.get(*cursor) {
                 raw.extend_from_slice(orig_name_bytes);
-            } else {
+                raw.extend_from_slice(b": ");
-                // Header not in original request (added by proxy) — use lowercase
+                raw.extend_from_slice(value.as_bytes());
-                raw.extend_from_slice(name_str.as_bytes());
+                raw.extend_from_slice(b"\r\n");
+                *cursor += 1;
+                emitted.insert(lower_name.clone());
             }
-            raw.extend_from_slice(b": ");
+        }
-            raw.extend_from_slice(value.as_bytes());
+    }
-            raw.extend_from_slice(b"\r\n");
+
+    // Append proxy-added headers (not present in the original request).
+    for name in headers.keys() {
+        let lower = name.as_str().to_ascii_lowercase();
+        if !emitted.contains(&lower) {
+            for value in headers.get_all(name) {
+                raw.extend_from_slice(name.as_str().as_bytes());
+                raw.extend_from_slice(b": ");
+                raw.extend_from_slice(value.as_bytes());
+                raw.extend_from_slice(b"\r\n");
+            }
+            emitted.insert(lower);
         }
     }
 

src-tauri/src/proxy/response_processor.rs

@@ -76,13 +76,29 @@ pub(crate) fn strip_entity_headers_for_rebuilt_body(headers: &mut HeaderMap) {
 }
 
 /// 读取响应体并在需要时解压，确保 headers 与返回 body 一致。
+///
+/// `body_timeout`: 整包超时。当非零时用 `tokio::time::timeout` 包住 `.bytes()` 调用，
+/// 防止上游发完响应头后卡住 body 导致请求永远挂住。
+/// 传入 `Duration::ZERO` 表示不启用超时（故障转移关闭时）。
 pub(crate) async fn read_decoded_body(
     response: ProxyResponse,
     tag: &str,
+    body_timeout: Duration,
 ) -> Result<(HeaderMap, http::StatusCode, Bytes), ProxyError> {
     let mut headers = response.headers().clone();
     let status = response.status();
-    let raw_bytes = response.bytes().await?;
+    let raw_bytes = if body_timeout.is_zero() {
+        response.bytes().await?
+    } else {
+        tokio::time::timeout(body_timeout, response.bytes())
+            .await
+            .map_err(|_| {
+                ProxyError::Timeout(format!(
+                    "响应体读取超时: {}s（上游发完响应头后 body 未到达）",
+                    body_timeout.as_secs()
+                ))
+            })??
+    };
 
     log::debug!(
         "[{tag}] 已接收上游响应体: status={}, bytes={}, headers={}",
@@ -184,7 +200,15 @@ pub async fn handle_non_streaming(
     state: &ProxyState,
     parser_config: &UsageParserConfig,
 ) -> Result<Response, ProxyError> {
-    let (response_headers, status, body_bytes) = read_decoded_body(response, ctx.tag).await?;
+    // 整包超时：仅在故障转移开启且配置值非零时生效
+    let body_timeout =
+        if ctx.app_config.auto_failover_enabled && ctx.app_config.non_streaming_timeout > 0 {
+            Duration::from_secs(ctx.app_config.non_streaming_timeout as u64)
+        } else {
+            Duration::ZERO
+        };
+    let (response_headers, status, body_bytes) =
+        read_decoded_body(response, ctx.tag, body_timeout).await?;
 
     log::debug!(
         "[{}] 上游响应体内容: {}",