Created by: NHZlX

We have upgraded the Paddle-TRT to support the model trained by paddleslim.

There are the following requirements for the model supported by paddleslim:

0. Normally, we only quant the backbone of the model.
1. the weight of the model should be quantized by channel_wise_abs_max.
2. add Compressor for conv and mud, and add AddQuantDequantPass for pool2d, elementwise_add, leaky_relu.