[REVIEW] Update cugraph_dgl to use the new FeatureStore

[VibhuJawa]

This PR updates cugraph_dgl to use the new FeatureStore (We see a drastic increase in fetching feature perf, 1900x even with host pytorch Tensors. I expect better perf with UVA tensors which we should be able to do with rmm's managed memory .

Speedup :

• 1900x faster get_feature_data
• 20x faster in Heterograph Sampling on obgn-mag

| Implimentation    | N_GPUS | Earlier Time Taken (s) | PR time taken (s) | SpeedUP | Batch Size |
|-------------------|--------|------------------------|-------------------|---------|------------|
| DGL Upstream CUDA | 1      | 0.26                   | NA                | NA      | 100_000    |
| Cugraph Local     | 1      | 2.175                  | 0.499             | 5x      | 100_000    |
| Cugraph Dask      | 1      | 113.511                | 5.20              | 21.73x  | 100_000    |
| Cugraph Dask      | 2      | 106.184                | 5.722             | 18x     | 100_000    |
| Cugraph Dask      | 4      | 102.991                | 6.302             | 16.19x  | 100_000    |

This PR also closes #3146 and #3039 and #3106

• Passes cugraph_dgl unit tests
• Add tests for dgl_uniform_sampler.py
• Verify cugraph_dgl works with obgn-products dataloader
  Verify cugraph_dgl works with ogbn-papers100M (To verify large scale scaling of the graphstore)
  Need bigger machine will do in a followup
• Add a dataloader based examples to verify E2E data loading works.

To decice:
Should we remove all the dead-code related to cugraph_service_store, cugraph_store etc from the cugraph repo ?

Benchmarks:

PR:

indices=np.random.randint(0,cugraph_g.num_nodes(),size=20_000)
%time output = cugraph_g.get_node_storage(key='feat',ntype='_N').fetch(indices)
CPU times: user 132 ms, sys: 263 ms, total: 395 ms
Wall time: 14 ms

Mainline:

indices=np.random.randint(0,cugraph_g.num_nodes(),size=20_000)
%time output = cugraph_g.get_node_storage(key='feat',ntype='_N').fetch(indices)
CPU times: user 132 ms, sys: 263 ms, total: 395 ms
Wall time: 27.8 s

[codecov-commenter]

Codecov Report

Base: 55.31% // Head: 55.60% // Increases project coverage by +0.29% 🎉

Coverage data is based on head (0e3cc82) compared to base (ea132d3).
Patch coverage: 93.51% of modified lines in pull request are covered.

Additional details and impacted files

@@               Coverage Diff                @@
##           branch-23.02    #3143      +/-   ##
================================================
+ Coverage         55.31%   55.60%   +0.29%     
================================================
  Files               148      143       -5     
  Lines              9423     9318     -105     
================================================
- Hits               5212     5181      -31     
+ Misses             4211     4137      -74     

Impacted Files	Coverage Δ
...graph/cugraph/gnn/dgl_extensions/utils/sampling.py	88.88% <87.50%> (+1.70%)	⬆️
.../cugraph/gnn/dgl_extensions/dgl_uniform_sampler.py	95.00% <95.00%> (ø)
...ugraph/cugraph/gnn/feature_storage/feat_storage.py	85.10% <100.00%> (ø)
python/pylibcugraph/pylibcugraph/testing/utils.py
...thon/pylibcugraph/pylibcugraph/testing/__init__.py
...n/pylibcugraph/pylibcugraph/utilities/api_tools.py
...pylibcugraph/pylibcugraph/experimental/__init__.py
python/pylibcugraph/pylibcugraph/__init__.py
python/pylibcugraph/pylibcugraph/_version.py

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[rlratzel]

Thanks, this is a large PR which I may have to go over again post-merge, but I did find a few minor things now which shouldn't hold up approval.

[alexbarghi-nv]

/merge