Revert chunk size to 8KB — larger sizes cause AWS deserialization errors

Keep 8KB CHUNK_SIZE (proven stable) but replace 10ms setTimeout delay
with a microtask break every 16 chunks. This avoids the AWS SDK
"Deserialization error: inspect {error}.\$response" while still
eliminating the ~1.25s/MB artificial delay from the old 10ms sleep.

src/utils/transcribeAWS.js

@@ -59,11 +59,16 @@ function convertToPCM(inputBuffer) {
 }
 
 async function* makeAudioStream(buffer) {
-  // Send chunks as fast as possible — AWS handles backpressure via HTTP/2 flow control.
-  // Larger chunks (32KB) reduce overhead and speed up streaming significantly.
-  var streamChunkSize = 32768; // 32 KB (AWS max per event frame)
-  for (var i = 0; i < buffer.length; i += streamChunkSize) {
-    yield { AudioEvent: { AudioChunk: buffer.slice(i, i + streamChunkSize) } };
+  // 8KB chunks — larger sizes cause AWS SDK deserialization errors
+  // ("to see the raw response, inspect the hidden field {error}.$response").
+  // No artificial delay between chunks; yield with a microtask break
+  // to avoid blocking the event loop without adding real latency.
+  for (var i = 0; i < buffer.length; i += CHUNK_SIZE) {
+    yield { AudioEvent: { AudioChunk: buffer.slice(i, i + CHUNK_SIZE) } };
+    // Microtask break every 16 chunks (~128KB) to keep event loop responsive
+    if ((i / CHUNK_SIZE) % 16 === 15) {
+      await new Promise(function(r) { setImmediate(r); });
+    }
   }
 }