[examples] Add Gemmini depthwise_conv examples

[cmx-Y]

Two Gemmini depthwise convolution examples have been added.

In dw_conv_2d_nchw_chw_i8.mlir, the -convert-linalg-to-loops option is used to convert linalg.depthwise_conv_2d_nchw_chw into loops for computation. In the future, a new pass may be added to enable its computation to leverage gemmini.tile_conv.

In dw-tile-conv.mlir, the multiple channels of the input tensor are split, and gemmini.tile_conv is invoked multiple times to compute the depthwise convolution. The correctness of this approach has been verified, although its performance might be relatively low and could be improved later.