Implementing tensor dot

[ariaghora]

Afaik, most of neural nets do not require tensor dot. However, wouldn't it be neat to have tensor dot implemented? First thing comes to my mind is treating tensor dot as batch matmul, so at least we can BLAS them.

Example of expected behavior: https://docs.scipy.org/doc/numpy-1.14.0/reference/generated/numpy.tensordot.html