fix the conv_2d batch mode

[bssrdf]

I think there is a bug in ggml_conv_2d which mixed up convolution results from different images when run in batch mode.
Specifically, this line

result = ggml_reshape_4d(ctx, result, im2col->ne[1], im2col->ne[2], a->ne[3], im2col->ne[3]); // [N, OC, OH, OW]

is supposed to reshape result to [N, OC, OH, OW]. However, values for a particular i of dimension N (batch size) are mixed from
j's.

This can be checked by making a stack of 2 input images each with identical but different values and observe the convolution mixed from both images.

The proper way to do this step is

result = ggml_reshape_4d(ctx, result, im2col->ne[1], im2col->ne[2], im2col->ne[3], a->ne[3]); // [ OC, N, OH, OW]    
result = ggml_cont(ctx, ggml_permute(ctx, result, 0, 1, 3, 2)); // [N, OC, OH, OW]