Fix incorrect types passed to Py_BuildValue and PyArg_ParseTupleAndKeywords

[QuLogic]

Always using i is broken, as that is 4 bytes, but Py_ssize_t and Py_uintptr_t are 8 bytes.

On little endian machines, this is usually fine, since the first 4 bytes (of 4-byte or 8-byte integers) will be the same, but on big-endian machines, a 4-byte integer corresponds with the last 4 bytes of the 8-byte integer

For small numbers on big-endian machines, it is thus likely that the result will be 0. If the value is over 32 bits, then the result will be wrong on little-endian machines as well.

Fixes #142

[QuLogic]

Here's a downstream build with this PR applied that works on all (little- and big-endian) architectures: https://src.fedoraproject.org/rpms/python-pyahocorasick/pull-request/3

[QuLogic]

There seems to be one more oddity at:

pyahocorasick/src/AutomatonItemsIter.c

Lines 254 to 262 in 88256f5

	#ifdef PY3K
	#ifdef AHOCORASICK_UNICODE
	"(u#O)", /*key*/ iter->buffer + 1, depth,
	#else
	"(y#O)", /*key*/ iter->buffer + 1, depth,
	#endif
	#else
	"(s#O)", /*key*/ iter->char_buffer + 1, depth,
	#endif

vs:

pyahocorasick/src/common.h

Lines 49 to 66 in 88256f5

	// setup supported character set
	#ifdef AHOCORASICK_UNICODE
	# if defined PEP393_UNICODE || defined Py_UNICODE_WIDE
	// Either Python uses UCS-4 or we don't know what Python uses,
	// but we use UCS-4
	# define TRIE_LETTER_TYPE uint32_t
	# define TRIE_LETTER_SIZE 4
	# else
	// Python use UCS-2
	# define TRIE_LETTER_TYPE uint16_t
	# define TRIE_LETTER_SIZE 2
	# define VARIABLE_LEN_CHARCODES 1
	# endif
	#else
	// only bytes are supported
	# define TRIE_LETTER_TYPE uint16_t
	# define TRIE_LETTER_SIZE 2
	#endif

where if AHOCORASICK_UNICODE is not defined, it states that it is const char * to Python, but the TRIE_LETTER_TYPE is uint16_t. I'm not sure if it's the former or latter that is wrong.