torch_tensor_to_array does not account for index ordering

[jwallwork23]

Running the tensor manipulation example proposed in #291, I notice that torch_tensor_to_array does not properly account for index ordering. The output of the example is currently

test 1
    Start 1: tensor_manipulation

1: Test command: /home/joe/software/FTorch/build/examples/0_Tensor/tensor_manipulation
1: Test timeout computed to be: 1500
1:  Contents of first input tensor:
1:  1  1  1
1:  1  1  1
1: [ CPUFloatType{2,3} ]
1:  Contents of second input tensor:
1:    1.00000000       2.00000000       3.00000000       4.00000000       5.00000000       6.00000000    
1:  Output:
1:    2.00000000       4.00000000       6.00000000       3.00000000       5.00000000       7.00000000    
1:  Tensor manipulation example ran successfully
1/1 Test #1: tensor_manipulation ..............   Passed    0.23 sec

100% tests passed, 0 tests failed out of 1

It takes a $2\times3$ tensor of ones and adds it to a $2\times3$ tensor containing the integers 1-6 that was created using the Fortran reshape intrinsic. The array generated by the reshape is printed as the "second input tensor". We do the addition using the overloaded + operator but then when we call torch_tensor_to_array, the output is not what we'd expect.

I think the issue is that torch_tensor_from_blob accounts for strides, whereas torch_tensor_to_blob doesn't. Therefore, we switch index orderings from Fortran to C, but then don't switch back.

[jwallwork23]

Closed as completed by #303.