Floating point precision error

math.huge does not unpack to the same value (in my machine, it in fact unpacks to a negative number).

I found what the issue was, it wasn't a precision error at all. You were deciding whether a lua_Number n represented a floating point value (float/double) instead of an integer through the comparison floor(n) != n. This fails when n is +-inf, since then floor(n) == n. When n is NaN, this also fails, since even tough floor(n) will return n, NaN fails to compare equal to any number, even itself. Your method also has the issue of not checking whether the floating point value could actually be represented by an int64_t, if it is integral in value.

I'm making a fork of the project, intending to send a pull request later.