opt: eval_network combine tensor_matvec and tensor_add

[mseminatore]

May be able to do the mul and add in a single call.

Alternatively, add the bias vector as part of the activation function