Compute AUC score as additional metric. If model has several output neurons with binary crossentropy loss, will use the average score.

## Examples

```
if (FALSE) { # reticulate::py_module_available("tensorflow")
y_true <- c(1,0,0,1,1,0,1,0,0) %>% matrix(ncol = 3)
y_pred <- c(0.9,0.05,0.05,0.9,0.05,0.05,0.9,0.05,0.05) %>% matrix(ncol = 3)
auc_metric <- auc_wrapper(3L, "binary_crossentropy")
auc_metric$update_state(y_true, y_pred)
auc_metric$result()
# add metric to a model
num_targets <- 4
model <- create_model_lstm_cnn(maxlen = 20,
layer_lstm = 8,
bal_acc = FALSE,
last_layer_activation = "sigmoid",
loss_fn = "binary_crossentropy",
layer_dense = c(8, num_targets))
auc_metric <- auc_wrapper(num_targets, loss = model$loss)
model %>% keras::compile(loss = model$loss,
optimizer = model$optimizer,
metrics = c(model$metrics, auc_metric))
}
```