Reading RK Number type is buggy

[fenisteel]

If you try to read large negative numbers, like -11 678 001, you got 1 062 063 823 instead. To fix this, replace the RKNum struct's number() method to this one:

func (r *RKNum) number() (intNum int64, floatNum float64, isFloat bool) {
	rk := helpers.BytesToUint32(r[:])
	isFloat = rk&0x02 == 0
	isMul := rk&0x01 == 1
	if isFloat {
		floatNum = math.Float64frombits(uint64(rk&0xfffffffc) << 32)
		if isMul {
			floatNum /= 100
		}
	} else {
		intNum32 := uint32(rk >> 2)
		if rk&0x80000000 > 0 {
			intNum32 = intNum32 | 0xC0000000
		}
		if isMul {
			intNum32 /= 100
		}
		intNum = int64(int32(intNum32))
	}
	return
}

The problem here, that there is an undocumented sign handling by Microsoft, that you must take into consideration. After you convert the number to an uint32, you must check the original number's leftmost bit, and if its one, then you must OR your number with 0xC0000000 to make the overflowing 32 bit uint into a signed 32 bit int. Then you can convert this into a 64bit signed int.