removeLayer() causes vector subscript out of range error

[MicheleCancilla]

Describe the bug Removing the last layer of Net causes vector subscript out of range debug assertion error in debug on Visual Studio. The error occurs on net_build.cpp: https://github.com/deephealthproject/eddl/blob/6759e7976f045e3cd10e43a3f23a9be8e4880562/src/net/net_build.cpp#L61

The last layer apparently has empty child vector, so layers[j]->child[0] causes out of range error.

To Reproduce Steps to reproduce the behavior:

1. Run onnx_export to create the "trained_model.onnx" file
2. Edit the file 3_onnx_import_net_from_file.cpp adding the removeLayer of last softmax:

   Net* net = import_net_from_onnx_file(path, DEV_CPU);
   removeLayer(net, "softmax6");
   // Build model
   build(net,
         rmsprop(0.01), // Optimizer
         {"softmax_cross_entropy"}, // Losses
         {"categorical_accuracy"}, // Metrics
         CS_GPU({1}, "low_mem"), // one GPU
         //CS_CPU(), // CPU with maximum threads availables
         false
   );

3. Debug the example onnx_import

[RParedesPalacios]

Here it is important to decide which layers are going to be the output of the network. If you remove the softmax but still pretend to build it with some loss and metrics associated... it has non sense

[RParedesPalacios]

Let me work on this and i will provide a way of doing

[RParedesPalacios]

Check 3_onnx_import_net_from_file.cpp in develop branch