fix: cuda memory allocation for cuda tests and fix step incrementing for LR update

Context

For Cuda test in Learning Module, the compiler requires casting float** into void** for cudaMalloc().
Fixes issue #13 (moved)

Also, as pointed in issue #12 (moved), LRScheduler' unit tests sometimes fails.
This error is due to incrementation of mStep: instead of (mStep++ < mSwitchStep) which increments after the comparison, the wanted behavior is to increment before the comparison, so ++mStep.

Modified files

• LRScheduler.hpp fix step incrementing for LR updating;
• Test_LRScheduler.cpp separate profiler testing for easier debug;
  Fix casting from float** to void** in the following files:
• Test_BCE.cpp;
• Test_MSE.cpp;
• Test_Accuracy.cpp;
• Test_Adam.cpp;
• Test_SGD.cpp;