[factorization] m4 macro for CUDA

[piyueh]

These lines of code in the m4 macro for CUDA will use $CUDA_DIR/lib in priority, if $CUDA_DIR/lib exists. However, on systems like Colonial One, $CUDA_DIR/lib and $CUDA_DIR/lib64 both exist, while only $CUDA_DIR/lib64 works. Making $CUDA_DIR/lib the first choice will stop configuration on systems like Colonial One.