Add support of const fused type memory views

[t20100]

This PR is related to #1772 and is an attempt at supporting const fused type memory views such as:

cimport cython
def sum(const cython.floating[:] array):
    cdef cython.floating sum = 0
    for v in array:
        sum += v
    return sum

Cython internals are new to me, so I don't know how far this is from a proper solution, but it's working fine from what I tested.
Let me know if it is worth spending more time on it and what can be improved.

One remark: In case of function signature like def test(const cython.floating[:] a, cython.foating[:] b) and mismatching input floating types, the exception is:
TypeError: No matching signature found
while without const it is: ValueError: Buffer dtype mismatch, expected 'float' but got 'double'

Closes #1772.

[scoder]

Thanks, I left some comments.

[t20100]

PR updated to use CConstOrVolatileType: change is now a single line.
More tests still need to be added.

[scoder]

Change looks perfect, but yes, more tests for const usage with (non-memoryview) fused types are needed. I can't currently find the word const anywhere in fused_types.pyx (probably because it came in later as a feature), so the combination of both features is very much undertested.

[t20100]

Added a few tests of const fused types usage without memory view.
Tell me if other use cases of const (and which one?) needs to be tested.

[scoder]

Thanks!

[lesteve]

Wow nice! Thanks a lot for this, this is something we needed in scikit-learn for some time. I'll try to get someone to try this out in scikit-learn and give feed-back.