*   Add delegate support for `QUANTIZED_16BIT_LSTM`.
    *   Converts hardswish subgraphs into atomic ops.
*   Add support for defaulting the value of `cycle_length` argument of
    `tf.data.Dataset.interleave` to the number of schedulable CPU cores.
*   `parallel_for`: Add converter for `MatrixDiag`.
*   Add `narrow_range` attribute to `QuantizeAndDequantizeV2` and V3.
*   Added new op: `tf.strings.unsorted_segment_join`.

   *  * If it is a message frame this will result in a single call to [FrameCallback.onReadMessage].
   *    If the message spans multiple frames, each interleaved control frame will result in a
   *    corresponding call to [FrameCallback].
   */
  @Throws(IOException::class)
  fun processNextFrame() {
    readHeader()
    if (isControlFrame) {
      readControlFrame()