Created by: songyouwei

This PR add Linear layer, a fully-connected linear transformation layer. Different from FC layer, Linear layer takes only one Tensor input.

The Linear layer uses matmul op while FC layer uses mul op, so they have different output shape.

• Linear multiplies input tensor with weight matrix and produces an output Tensor of shape [N, *, output_dim], where N is batch size and * means any number of additional dimensions.
• FC can take one or multiple Tensor as its inputs, and output shape is[N, size], where N is batch size.