Batched tagger example

This is for CoNLL-2003 datasets for NER tasks. Tagger example is taken from here:
(https://github.com/clab/dynet/tree/master/examples/tagger)

Speed of BiLSTM tagger example on CPU: 6.033s
Speed of BiLSTM tagger example on GPU: 37.4s (for GPU I used the dynet-gpu option)

Also the nvprof profiler is getting killed after 1 optimization step.

Can you please suggest the correct way of using GPU in dynet ?

Start by reproducing the DyNet tagger example from examples/tagger with the dynet-gpu option on the reported CoNLL-2003 workload. Compare the CPU and GPU timings and check the nvprof behavior after one optimization step; the issue is complete only when the correct GPU usage and profiler failure are explained.

Written by the indexing model from the issue text.