Fix static-size FFI type aliases

Several type aliases with statically sized names (e.g. `npy_uint64`)
were being aliased to platform-dependent types like
`::std::os::raw::c_long`.  Most of these were barely used, so it seems
like they didn't cause problems before, but now with `npy_uint64` being
used in certain structs, which are the target of pointer punning, the
mismatch in sizes became visible.  For example, `npy_uint64` was
previously typedef'd to `c_ulong`, which is 64 bits on 64-bit platforms
and on all Unix-likes, but 32 bits on Windows 32-bit.  This caused
inaccurate reads through pointers to structs containing them, eventually
resulting in attempts to derefence null pointers.

src/npyffi/types.rs

@@ -19,15 +19,15 @@ pub type npy_long = c_long;
 pub type npy_float = f32;
 pub type npy_double = f64;
 pub type npy_hash_t = Py_hash_t;
-pub type npy_int64 = c_long;
-pub type npy_uint64 = c_ulong;
-pub type npy_int32 = c_int;
-pub type npy_uint32 = c_uint;
+pub type npy_int64 = i64;
+pub type npy_uint64 = u64;
+pub type npy_int32 = i32;
+pub type npy_uint32 = u32;
 pub type npy_ucs4 = c_uint;
-pub type npy_int16 = c_short;
-pub type npy_uint16 = c_ushort;
-pub type npy_int8 = c_char;
-pub type npy_uint8 = c_uchar;
+pub type npy_int16 = i16;
+pub type npy_uint16 = u16;
+pub type npy_int8 = i8;
+pub type npy_uint8 = u8;
 pub type npy_float64 = f64;
 pub type npy_complex128 = npy_cdouble;
 pub type npy_float32 = f32;