qa not calculated correctly in backpropagation

[andreicscs]

nn.h
row 419: float qa = dactf(a, NN_ACT);

"qa" the output of the activation function's derivative should be derived from the pre-activation output (o or z) instead of directly using the activated output. This is crucial because the derivative of the activation function must be computed with respect to the input to that function, which is the pre-activation value.

therefore in the current code the chainrule is not applied correctly