perf(eio): stream compressed polling responses

Pipe gzip/deflate output directly to the polling HTTP response instead of
buffering compressed chunks in memory before sending them.

This reduces memory usage for large compressed polling payloads while keeping
the existing compression threshold and content negotiation behavior.

See also: d11e17c (engine.io@1.6.0, Nov 2015)

Author: darrachequesne

packages/engine.io/lib/transports/polling.ts

@@ -314,40 +314,29 @@ export class Polling extends Transport {
       return;
     }
 
-    this.compress(data, encoding, (err, data) => {
-      if (err) {
-        this.res.writeHead(500);
-        this.res.end();
-        callback(err);
-        return;
-      }
+    debug("compressing");
 
-      headers["Content-Encoding"] = encoding;
-      respond(data);
-    });
-  }
+    headers["Content-Encoding"] = encoding;
+    this.res.writeHead(200, this.headers(this.req, headers));
 
-  /**
-   * Compresses data.
-   *
-   * @private
-   */
-  private compress(data, encoding, callback) {
-    debug("compressing");
+    const stream = compressionMethods[encoding](this.httpCompression);
 
-    const buffers = [];
-    let nread = 0;
+    let isErrored = false;
 
-    compressionMethods[encoding](this.httpCompression)
-      .on("error", callback)
-      .on("data", function (chunk) {
-        buffers.push(chunk);
-        nread += chunk.length;
-      })
-      .on("end", function () {
-        callback(null, Buffer.concat(buffers, nread));
-      })
-      .end(data);
+    stream.on("error", (err) => {
+      isErrored = true;
+      this.res.end();
+      callback(err);
+    });
+
+    this.res.once("finish", () => {
+      if (!isErrored) {
+        callback();
+      }
+    });
+
+    stream.pipe(this.res);
+    stream.end(data);
   }
 
   /**