add openMP pragma to warp, hflip (non-inplace) and vflip (non-inplace).

[jonathantompson]

The tests all pass (although tests in image are pretty bare-bones).

Someone with more OpenMP experience please check it (i.e. variables declared on the stack (in the loop) are made thread-private by default right?).

For bicubic warp I get about 10x speed up using 12 cores on a 256x256 image.

[soumith]

Thanks Jono.