Use Numpy axis normalizations where possible

[bryevdv]

issues: fixes #412

@magnatelee I think this is smaller in scope than you have imagined, but I am not sure much more is advised. Some notes:

• array.py, sort.py, linalg.py

  I have applied the Numpy normalization functions in places where it seemed to make sense. Most locations were just passing values other functions.

• fft.py

  Most locations just do computed_axis = (axis,) if axis is not None else None which does not seem worth defining a new function for. One place calls a _sanitize_user_axes but that function does unique conversion that takes shape into account.

• deferred.py

  A few uses here, but handles comepletely differently, e.g. via result.project

• ufunc.py, eager.py

  The few uses here just pass on to other functions

• thunk.py

  no code, just signatures