make encoding detection code a bit faster if the first two source file lines are longer than 250 bytes

Get rid of obsolete Python cruft.

cython-mode.el: don't highlight builtins when they're part of varname

simplify test for GCC >= 2.95 for branch prediction

document memoryview attributes better

--HG--
extra : amend_source : 9ae2e52935ae16af23c35b6329d53e559b7c39d0

Fixed typo

(I know, I know. It was just too visible.)

To avoid mismatches keywords should be checked at symbol, not word
boundaries, otherwise they'll ignore underscores (is_*float*).

apply freelist type check also when *adding* dead objects to the freelist to make sure their eventual cleanup works correctly

--HG--
extra : amend_source : 6c9be6e15396919e08090d9d12557c4255fd8a3c

revert to using malloc/free for data buffer allocation in case external code wants to fiddle with it

make it easier for the C compiler to prune some useless error checking code in cython.array constructor

--HG--
extra : amend_source : 4d5d9dc96421710652afe8614c9f32db26cc88c1

--HG--
extra : amend_source : 685e555a2102e496280287b3b7605d67d694ff40

use CPython heap memory allocation for memory views where possible (outside of nogil sections) as it tends to be faster than plain malloc()