Remove and Matrix::convExpand and Matrix::convShrink and Refactor BlockExpandLayer

[hedaoyuan]

1. Remove the Matrix::convExpand and Matrix::convShrink code, it has been re-implemented in the Function.
2. Reconstruction of this BlockExpandLayer with the implementation of Function.

[hedaoyuan]

• Image Expand Function
  Need to implement an image expand Function, Converts an NCHW format image input data into an NST format sequence data. Where N is batch size, S is the length of the sequence after each image is expanded, T is the size of each time step in the sequence.

• Modify Im2ColFunctor implementation
  The input and output of Matrix::convExpand and Im2ColFunctor are in the following format,

imData = [input_channels, input_height, input_width]
colData = [input_channels, filter_height, filter_width, output_height, output_width]

The output in the Image Expand Function requires [output_height, output_width, input_channels, filter_height, filter_width] format, this can be convenient for reshape into [S, T] format, where S is output_height * output_width, and T is input_channels * filter_height * filter_width. Otherwise, you will need to do a transpose operation on the result of Im2ColFunctor. transpose in BlockExpandLayer