1. make sure they are tuples
2. was checking if the max axis was > available dim, should have been
>= available dim

update unit tests to use new tensordot function, remove infer_shape tests for tensordot and tensordotgrad

the closing """ is ignored because of the #

ygrad still requires a transpose.

-call numpy.dot for inputs with dim == 0 or dim > 2
-add gradient calcs for various cases
-add infer_shape for various cases
-update docstring

Add sort method and tensor.argsort function

Py2.4

Use other impl. of OrderedSet to prevent mem. leak

Previous implementation resulted in memory not being freed,
as exhibited in particular on theano/sandbox/cuda/tests/test_memory.py.

The version for older Python, as well as the new one, appear to work
correctly on that test.