fix: 优化上行数据处理逻辑，增加取消上行reader的处理，改进缓冲管理

Author: cmliu

_worker.js

@@ -664,15 +664,23 @@ async function 处理叉HTTP请求(request, yourUUID, 反代上下文 = {}) {
 		const 搬运Promise = 上行合包器.readable.pipeTo(socket.writable, { signal: abortController.signal });
 		void 搬运Promise.catch(() => { });
 		const 上行reader = request.body.getReader();
+		const 取消上行reader = () => {
+			try { 上行reader.cancel(abortController.signal.reason).catch(() => { }); } catch (e) { }
+		};
+		abortController.signal.addEventListener('abort', 取消上行reader, { once: true });
 		try {
-			while (true) {
-				const { done, value } = await 上行reader.read();
-				if (done) break;
-				if (value?.byteLength) await 上行合包器.写入(value);
+			try {
+				while (true) {
+					const { done, value } = await 上行reader.read();
+					if (done) break;
+					if (value?.byteLength) await 上行合包器.写入(value);
+				}
+			} finally {
+				abortController.signal.removeEventListener('abort', 取消上行reader);
+				try { 上行reader.releaseLock() } catch (e) { }
 			}
-			await 上行合包器.结束();
 		} finally {
-			try { 上行reader.releaseLock() } catch (e) { }
+			try { await 上行合包器.结束() } catch (e) { }
 		}
 		await 搬运Promise;
 	})();
@@ -2592,7 +2600,8 @@ function 创建上行Grain合包流(目标字节 = 上行合包目标字节) {
 		? new IdentityTransformStream()
 		: new TransformStream();
 	const writer = identity.writable.getWriter();
-	let 缓冲 = new Uint8Array(0);
+	const 缓冲 = new Uint8Array(目标字节);
+	let 缓冲长度 = 0;
 	let 定时器 = null;
 	let 在途写 = null;
 
@@ -2603,12 +2612,17 @@ function 创建上行Grain合包流(目标字节 = 上行合包目标字节) {
 		}
 	};
 
+	const 串行写 = async (chunk) => {
+		if (在途写) await 在途写;
+		在途写 = writer.write(chunk);
+		try { await 在途写 } finally { 在途写 = null; }
+	};
+
 	const 冲刷 = async () => {
-		if (缓冲.byteLength) {
-			const chunk = 缓冲;
-			缓冲 = new Uint8Array(0);
-			在途写 = writer.write(chunk);
-			try { await 在途写 } finally { 在途写 = null; }
+		if (缓冲长度) {
+			const chunk = 缓冲.slice(0, 缓冲长度);
+			缓冲长度 = 0;
+			await 串行写(chunk);
 		}
 	};
 
@@ -2623,27 +2637,23 @@ function 创建上行Grain合包流(目标字节 = 上行合包目标字节) {
 	return {
 		readable: identity.readable,
 		写入: async (chunk) => {
-			if (在途写) await 在途写;
 			const data = 数据转Uint8Array(chunk);
 			if (!data.byteLength) return;
-			if (缓冲.byteLength === 0 && data.byteLength >= 目标字节) {
+			if (缓冲长度 === 0 && data.byteLength >= 目标字节) {
 				清理定时器();
-				await writer.write(data);
+				await 串行写(data);
 				return;
 			}
-			const 合并后 = 缓冲.byteLength + data.byteLength;
-			if (合并后 >= 目标字节) {
-				const output = new Uint8Array(合并后);
-				output.set(缓冲, 0);
-				output.set(data, 缓冲.byteLength);
-				缓冲 = new Uint8Array(0);
+			if (缓冲长度 + data.byteLength >= 目标字节) {
+				const output = new Uint8Array(缓冲长度 + data.byteLength);
+				output.set(缓冲.subarray(0, 缓冲长度), 0);
+				output.set(data, 缓冲长度);
+				缓冲长度 = 0;
 				清理定时器();
-				await writer.write(output);
+				await 串行写(output);
 			} else {
-				const output = new Uint8Array(合并后);
-				output.set(缓冲, 0);
-				output.set(data, 缓冲.byteLength);
-				缓冲 = output;
+				缓冲.set(data, 缓冲长度);
+				缓冲长度 += data.byteLength;
 				启动定时器();
 			}
 		},