Tensorflow bridge memory leak.

I was using decord to load videos in batches for tensorflow and was having a serious GPU memory leak (after several passes, I ran out of memory). Spend some time to figure out that the issue was with setting `decord.bridge.set_bridge('tensorflow')` and replaced it with loading batchs with `vr.get_batch(indexes).asnumpy()` and converting them with `tf.constant()`.