Support variable-length input feature for convolution operation.

[qingqing01]

Some applications, such as OCR, image segmentation, detection, speech, need to support variable-length input feature. Now, the API only supports fixed-length 2D or 3D feature and expands them to a vector as input.

For the 2-D convolution operation, each sample in one mini-batch must have the similarly size in PaddlePaddle now. But, it supports variable-dimension feature across mini-batch. We'd like to support this feature.

• Expose these interfaces for Arguments in swig firstly.
• Then support this feature in Python API.