Fix back propagation through time for both reference and CPU.

tmva/tmva/inc/TMVA/DNN/Architectures/Cpu.h

tmva/tmva/inc/TMVA/DNN/Architectures/Cpu/CpuMatrix.h

tmva/tmva/inc/TMVA/DNN/RNN/RNNLayer.h

tmva/tmva/src/DNN/Architectures/Cpu/Arithmetic.cxx

tmva/tmva/src/DNN/Architectures/Cpu/RecurrentPropagation.cxx

.../src/DNN/Architectures/Reference/RecurrentPropagation.cxx

tmva/tmva/src/MethodDL.cxx

tmva/tmva/test/DNN/RNN/CMakeLists.txt

tmva/tmva/test/DNN/RNN/TestRecurrentBackpropagation.cxx

tmva/tmva/test/DNN/RNN/TestRecurrentBackpropagation.h

tmva/tmva/test/DNN/RNN/TestRecurrentBackpropagationCpu.cxx