Feature Quantization [5/5]: Distributed Integration

[jchmura-sc]

Purpose
The purpose of this PR is to integrate feature quantization into our distributed data, loading, partioning, and sampling setup:

• graph partitioner partitions quantized node feature using node partion book
• neighbor samplers fetch quantized node features over rpc during subgraph hydration
• dataloaders dequantize/reconstruct node feature during collate_fn on gpu
• distributed dataset takes quantization metadata (to pass down to dataloaders) and new quantized node feature store

Context

• Feature Quantization [1/5]: Compile new protos #733
• Feature Quantization [2/5]: Core Quantization Ops #734
• Feature Quantization [3/5]: Preprocessor (write-side) Updates #735
• Feature Quantization [4/5]: Read updated TFRecords #736
• Feature Quantization [5/5]: Distributed Integration #737

Updated Changelog.md? NO

Ready for code review?: NO