auto check lapack libs

paddle.init() default use env

Expose Inference in Python V2 API.

Adjust float precision in img_conv2 for unit test

change to lapacke

modify img_conv2 unit test

remove cmake python output

Remove the feenableexcept in initPaddle.

force delete packages before unit test

fix gpu_img_conv2 float precision issue

using one thread to dump sequential output

'numpy.ndarray' object has no attribute 'tobytes'

run python tests in virtualenv