Add Tensor Format option to DLTensor

[ryanolson]

As a follow-on to Issue #34, I propose the addition of describing the tensor format for the underlying data in the DLTensor struct.

This option would consume a single byte with one of two possible values: kDLRowMajor or kDLColumnMajor.

[ryanolson]

Upon further thought, one can simple infer based on the strides if the underlying format is row/column major.

Unfortunately, that leaves some ambiguity in the reading of the definition of strides which is allowed to be NULL.

  /*!
   * \brief strides of the tensor,
   *  can be NULL, indicating tensor is compact.
   */
  int64_t* strides;

https://github.com/dmlc/dlpack/blob/master/include/dlpack/dlpack.h#L140-L144

It is important to be specific about the definition of a NULL value for strides. I don't have a preference here, but we should explicit state that a NULL value indicates a compact row-major (or column-major) tensor.

>>> c = np.empty((2,3,224,224), dtype=np.float32)
>>> c.shape
(2, 3, 224, 224)
>>> c.strides
(602112, 200704, 896, 4)
>>> f = np.empty((2,3,224,224), dtype=np.float32, order='f')
>>> f.shape
(2, 3, 224, 224)
>>> f.strides
(4, 8, 24, 5376)

[tqchen]

It is row-major by default, a PR to clarify things is more than welcomed