[DistPartPipeline] canonical etypes are used in `edge_feat.dgl` instead of etypes

[Rhett-Ying]

🐛 Bug

anyone has verified the partition outputs with edge feature works for distributed train?

'paper:cites:paper/count' is used as key of edge_feat.dgl while dgl.distributed.load_partition() supports etype only for now.

What's worse, the test code below does not take effect at all. key is not defined correctly. neither part_data. https://github.com/dmlc/dgl/blob/eae6ce2a4f44e81eda3b876ef50fc045926912f6/tests/tools/test_dist_part.py#L271-L281

This issue is introduced by https://github.com/dmlc/dgl/pull/4623

To Reproduce

Steps to reproduce the behavior: python3 -m pytest tests/tools/test_dist_part.py::test_part_pipeline

Expected behavior

Environment

• DGL Version (e.g., 1.0): master
• Backend Library & Version (e.g., PyTorch 0.4.1, MXNet/Gluon 1.3):
• OS (e.g., Linux):
• How you installed DGL (conda, pip, source):
• Build command you used (if compiling from source):
• Python version:
• CUDA/cuDNN version (if applicable):
• GPU models and configuration (e.g. V100):
• Any other relevant information:

Additional context