[Distributed] Specify the graph format for distributed training

[zheng-da]

Description

Previously, the graph structure in each partition is stored as CSC. When users call APIs such as find_edges and out_degrees, DGL generates corresponding graph format for these APIs. In the setting of distributed training, the graph structure is shared among all trainers and servers. If an API triggers the construction of another graph format, the graph structure of the new format is stored in local memory. If every trainer and server in a machine does the same thing, a graph will be replicated by many times in the machine.

This PR allows users to specify the graph formats during the launch time. Once the graph format is constructed, any graph API can no longer construct a new graph format during the runtime. If a graph format is required but not created during the launch time, an error is reported.

Checklist

Please feel free to remove inapplicable items for your PR.

• The PR title starts with [$CATEGORY] (such as [NN], [Model], [Doc], [Feature]])
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage
• Code is well-documented
• To the my best knowledge, examples are either not affected by this change,
  or have been fixed to be compatible with this change
• Related issue is referred in this PR
• If the PR is for a new model/paper, I've updated the example index here.

Changes

[dgl-bot]

To trigger regression tests:

• @dgl-bot run [instance-type] [which tests] [compare-with-branch];
  For example: @dgl-bot run g4dn.4xlarge all dmlc/master or @dgl-bot run c5.9xlarge kernel,api dmlc/master

[classicsong]

Overall LGTM