Use tensor as the bottom implementation and support plain scalar.
Note: DO NOT support the backward for ReduceSumAll function.

1. Initialize and new scalar tensor.
2. Get and set scalar tensor.
3. Add ReduceSum test on scalar tensor.
4. Little bugs fixed.

Fix the memory pool bug in SetDevice function.

Update Compare.cpp.

Bug fixed in Clang compiler.

Update ShowMemUsage function in XMem.cpp.

Update Compare.cpp.

1. Move the SetAscendingOrder functions from XTensor.*  to core/utilities/SetAscendingOder.*
2. Rename XElement.* files to XCall.*

1. Move the New/Initial/Del Tensor functions from XTensor.* to XElement.*
2. Move IsSameShaped functions from XTensor.* to core/shape/IsSameShaped.*
3. Move CheckData functions from XTensor.* to core/utilities/CheckData.*

# Conflicts:
#	source/tensor/XTensor.cpp
#	source/tensor/core/getandset/OnehotAndIndex.cpp
#	source/tensor/core/getandset/OnehotAndIndex.cu

When we do the "=" operation, the link will be copy to the new tensor instead of replace it.

2. fix the bug of onehotandindex
3. modify the data initilization of transformer
4. add the DumpFormat (need reimplement)