0-dimensional arrays

[alimanfoo]

This PR provides tests and implementation of support for 0-dimensional arrays. This builds off work done in #154, however I needed to add special case implementations of __getitem__ and __setitem__ to the Array class to get indexing working with both empty tuple and ellipsis.

[alimanfoo]

OK, I think this is good to go.

[jakirkham]

Minor point. This seems a bit spaced out. Not sure if that was intentional or not.

[alimanfoo]

Behaviour should be consistent with h5py and numpy, e.g.:

>>> import zarr
>>> z = zarr.zeros(())
>>> z
<zarr.core.Array () float64>
>>> z[...]
array(0.0)
>>> z[()]
0.0
>>> z[...] = 42
>>> z[...]
array(42.0)
>>> z[()] = 43
>>> z[()]
43.0
>>> z[:]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/aliman/src/github/alimanfoo/zarr/zarr/core.py", line 451, in __getitem__
    return self._getitem_zd(item)
  File "/home/aliman/src/github/alimanfoo/zarr/zarr/core.py", line 460, in _getitem_zd
    raise IndexError('too many indices for array')
IndexError: too many indices for array

[jakirkham]

Would it be worth checking nbytes as well or is that too far afield?

[jakirkham]

Does _cdata_shape need to be updated?

[alimanfoo]

Thanks @jakirkham, have pushed some tweaks to tests and properties.

[alimanfoo]

Also can now set scalar on group:

>>> import zarr
>>> g = zarr.group()
>>> g['foo'] = 42
>>> g['foo']
<zarr.core.Array '/foo' () int64>
>>> g['foo'][()]
42
>>> g['bar/baz/qux'] = 123.4
>>> g['bar/baz/qux']
<zarr.core.Array '/bar/baz/qux' () float64>
>>> g['bar/baz/qux'][()]
123.40000000000001

[jakirkham]

LGTM

[joshmoore]

As a reminder to future selves, JuliaIO/Zarr.jl#117