fix improper integer type on systems with non-32-bit int

[mrpapersonic]

generally the parameter here should be utf8proc_int32_t anyway, but changing public facing APIs for something like this isn't a very good idea. maybe this should be changed for the next major bump in the libtool version

[stevengj]

seqindex_write_char_decomposed is not part of the API, so we can just change the parameter.

[mrpapersonic]

utf8proc_decompose_char (the function this patch edits) is though. the most sensible change would be to edit last_boundclass to be 32-bit, but that breaks ABI, though I suppose it doesn't matter for the 0.0001% of systems that actually deal with 16-bit or 64-bit int.

[stevengj]

I think we can just change the utf8proc_decompose_char API to declare it as int32. ILP64 systems are vanishingly rare these days, and 16-bit is probably mainly embedded systems where I doubt people are using utf8proc as a shared library.