Cleanup of sparse/basic.py

Fix defaults

fixed debugging and warning messages (grammar and semantics) for cudandarray

previously, it stored the initial value of shared variables in defaults
this wasted a lot of memory
now it does not

Minor stuff

ndarray

Similar lists were added recently in elemwise.py: it is better to re-use
existing ones.

Fixed infinite canonizer loop with NaN constants

Update internal doc for release

Minor stuff

Comparing strings with 'is' is dangerous, as there is no guarantee that
Python will store similar strings into the same memory space.

In filter, raise TypeError or ValueError.