Refactor DGLKE

[menjarleev]

Issue #, if available:

• Currently only test TransE_l1 with wn18. Other models need to be tested in the future.

Description of changes:

• refactor entire DGLKE package to enable further extension.
• Package can are divided into: Model, Dataset, Dataloader.
• Dataset download benchmark dataset(FB15k) from web and split it into train, eval, test datasets. These datasets can be further partitioned into chunks to support single machine multi-GPU training.
• Dataloader sample data from partition dataset and feed them to model.
• A model consists of an encoder and (possibly multiple) decoder. The encoder fetch samples from dataloader, and output embeddings of graph based on the sliced indices. The decoder use embeddings from encoder to produce error and update parameters of model or generate evaluation results.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[zheng-da]

can you PR to the ng branch? v0.2 branch is used for releasing code.