Fix bug in residual_block

[reshalfahsi]

Hello sir,

In your paper, it is reported that the NanoNet-C has parameters of 36K. Yet, when I try to replicate the model in PyTorch, available online on my repository, the number of parameters doesn't match up. Instead, it has 43K parameters. After carefully reading your code, I found an error in the residual_block function. However, I have fixed the error summarized in this pull request. Thank you.