TENSOR: Indexing is complicated.

[ntjohnson1]

Our indexing always makes my head hurt because of the variety of paths through the logic but I think we are getting simpler/more robust. Adding this as a single commit so @DeepBlockDeepak can cherry-pick if desired to test if this unblocks the tensor tutorial work. Otherwise we can merge and iterate. See commit for the couple of changes I made. I updated our tests to cover the failure modes you described but let me know if you think I need to add something additional.

Interface Note: MATLAB uses a lot of column vectors to resolve ambiguities but those are generally less common in python since numpy defaults to a more generic (length,) array. My preference is to take a list or effective row vector for linear indices and force people trying to use subscripts to always to provide a matrix np.array([[0,1,2]]) even if the nesting isn't generically necessary.

For the exact motivation, here is the behavior off the branch which I believe aligns mostly with expectations.

>>> import pyttb as ttb
>>> import numpy as np
>>> X = ttb.tensor.from_data(np.arange(1,25),(3,4,2,1))
>>> S = [[0,0,0,0],[2,3,1,0]]
>>> X[S]
Traceback (most recent call last):
  File "<path>\Anaconda3\envs\pyttb\lib\site-packages\IPython\core\interactiveshell.py", line 3505, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-6-7b0c9de99a5d>", line 1, in <module>
    X[S]
  File "<path>\pyttb\pyttb\tensor.py", line 1573, in __getitem__
    assert False, "Invalid use of tensor getitem"
AssertionError: Invalid use of tensor getitem
>>> S = np.array([[0,0,0,0],[1,1,1,0],[2,3,1,0]])
>>> X[S]
array([ 1, 17, 24])
>>> S = [0,5,6,22]
>>> X[S]
array([ 1,  6,  7, 23])
>>> X[np.array(S)]
array([ 1,  6,  7, 23])