No endianness translation for GPU IPv4 lookup kernel

[Junhyun]

The CPU version of the code transforms IPv4 dest addr to LE using ntohl before lookup.
GPU kernel should (AFAIK, NVIDIA GPU kernels operate in LE as well), but does not, at the moment.
It is trivial, but crucial to the correctness of IPlookup kernel.
A simple 4-byte reordering would do the trick.

• Simply swap bytes in GPU kernels.