* Develop a fake dequantized op for fixed-point quantization training framework.

* Add the missing file.