test=develop
* Add a fake_quantize_op, which quantize an input tensor to a tensor with lower bits.
This is part of big move of v2 codes to legacy sub dir.
