Mygrad

Adds convenient overrides and mirrors to non-differentiable numpy functions and (non-complex) dtypes.

- allclose
- bincount
- can_cast
- copyto
- may_share_memory
- min_scalar_type
- result_type
- shares_memory
- shape

Implements `Tensor.__array_function__`.

Common numpy functions now have "drop-in" autodiff, when they operate on a tensor:

python
>>> x = mg.tensor([2., -1., 0.])
>>> numpy.einsum("i,i", x, x).backward()
>>> x.grad
array([ 4., -2., 0.])

Implements `__array_ufunc__` to override/dispatch NumPy ufuncs via `mygrad.Tensor`

See [changelog](https://mygrad.readthedocs.io/en/latest/changes.html)