Carray indexing return inconsistency with numpy

[phizaz]

An example is:

a = bcolz.carray(np.zeros([10, 1]))
a[0] # 0.0

While in numpy:

a = np.zeros([10, 1])
a[0] # array([0.])

This inconsistency causes a lot of problem on handling them transparently.

When the second dimension is larger, this problem vanishes:

a = bcolz.carray(np.zeros([10, 2]))
a[0] # array([0., 0.])