Drain all buffered ACKs per I/O loop iteration

Previously, tryReceiveAcks() called tryReceiveFrame() exactly once
per I/O loop iteration. In the DRAINING state (all batches sent, ACKs
still pending), the I/O thread sleeps 10 ms after each call. This meant
each in-flight frame added ~10 ms to flush latency: N in-flight frames
incurred ~N x 10 ms overhead, even when all ACKs were already sitting in
the TCP receive buffer.

Fix by looping tryReceiveFrame() until it returns false, draining all
buffered ACKs in a single pass. The 10 ms sleep is now reached only when
no more data is available, which is the correct condition for avoiding a
busy loop.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: mtopolnik

core/src/main/java/io/questdb/client/cutlass/qwp/client/WebSocketSendQueue.java

@@ -555,7 +555,9 @@ private void sendBatch(MicrobatchBuffer batch) {
      */
     private void tryReceiveAcks() {
         try {
-            client.tryReceiveFrame(responseHandler);
+            while (client.tryReceiveFrame(responseHandler)) {
+                // Drain all buffered ACKs before returning to the I/O loop.
+            }
         } catch (Exception e) {
             if (running) {
                 LOG.error("Error receiving response: {}", e.getMessage());