enhancements buffer syntax

It's been pretty much decided on to have dtype before ndim.

1. cdef ndarray[float, 2] arr = x
2. cdef ndarray<float, 2> arr = x Problem: Want to reserve this for wrapping C++ templates
3. cdef ndarray(float, 2) arr = x Problem: Looks like constructor
4. cdef ndarray arr[float, 2] = x Problem: Want to reserve this for C arrays
5. cdef buf(float, 2) arr = x Problem: See explicit syntax below.
6. A with statement without assigning to variable, like this:

with buffer(arr, float, 2):
    arr[10, 10] = 3.2

A bit verbose for many typical users though.

7. A with statement with assignment to a variable. See "explicit syntax" below.

Comments:

• None are liked...

A natural syntax would be 5. above or perhaps something like

#!python
with get_buffer(arr, float, 2) as buf:
    buf[2, 3] = 10

The main problem with these kind of syntax choices is that one would often like to use the buffer interface in conjunction with higher-level operations. NumPy example:

#!python
cdef ndarray[float, 2] arr = zeros((10, 10), dtype=float)
arr += 10.3
for i in range(10): arr[i, i] = 0
arr *= 10
for i in range(10): arr[10-i, 10-i] = 0

and so on.

Having to switch between "arr" and "buf" depending on the usecase is not going to be user-friendly:

#!python
cdef ndarray arr = zeros((10, 10), dtype=float)
with get_buffer(arr, float, 2) as buf:
    arr += 10.3
    for i in range(10): buf[i, i] = 0
    arr *= 10
    for i in range(10): buf[10-i, 10-i] = 0