Fix 32-bit arm int issue.

For (some?) 32-bit arm platforms, int32_t is actually of type long int, not int. We have a couple places where we assume extracting a bit pattern from a float is type int, when we should use int32_t instead. The discrepancy in types causes some packet functions to fail.

I'm not sure how prevalent this issue is, but these changes at least fix it for #2412 (closed).

Fixes #2412 (closed).