Flux Klein 9B in INT8 for 2x inference gains on 30-Series GPUs.

To be used with https://github.com/BobJohnson24/ComfyUI-Flux2-INT8

Comparison against FP8:

<table>
  <tr>
    <th>FP8</th>
    <th>INT8 Prequantized (this repo)</th>
    <th>INT8 dynamically loaded from bf16</th>
  </tr>
  <tr>
    <td><img src="FP8Grid.jpg" width="500" height="500"></td>
    <td><img src="INT8Grid_prequantized.jpg" width="500" height="500"></td>
    <td><img src="INT8Grid_dynamic.jpg" width="500" height="500"></td>
  </tr>
</table>