Auto shape for Web Service

support cuda10.1 10.2 TensorRT and fix some bug

support multiple devices(cpu/gpu/trt/xpu/arm) in pipeline mode

Develop

update xpu l3 cache size;

fix python bug and add java example

[WIP]support arm and xpu

Update channel_acc_size

fix compile doc for trt

fix param name

2.0 develop first commit

Update pip version

Update  pip install paddle-serving-app==0.2.0