new: adjust DGL to ArangoDB interface for increased accessibility

[aMahanna]

• Closes adjust DGL to ArangoDB interface for increased accessibility #23
• Minor housekeeping
• Drop "official" 3.6 support

Demo

ArangoDB to DGL use cases

dgl_karate_graph = KarateClubDataset()[0]

# Use Case 1: Let ADBDGL_Adapter handle the graph creation
adbdgl_adapter.dgl_to_arangodb("Karate", dgl_karate_graph)

# Use Case 2: Create the graph prior to ArangoDB to DGL call
edge_definitions = [...]
db.create_graph("Karate", edge_definitions, smart = True, disjoint = True, ...)
adbdgl_adapter.dgl_to_arangodb("Karate", dgl_karate_graph)

# Use Case 3: Overwrite the graph due to edge definition modifications
new_dgl_karate_graph = heterograph({("KarateMember", "to", "KarateMember"): (tensor([0, 1]), tensor([1, 2]))})
adbdgl_adapter.dgl_to_arangodb("Karate", new_dgl_karate_graph, overwrite_graph=True)

Keyword arguments for import_bulk

Users can now specify parameters for the import_bulk() API of python-arango via kwargs

adbdgl_adapter.dgl_to_arangodb("Karate", dgl_karate_graph, sync=True, on_duplicate="replace", batch_size=50)

[cw00dw0rd]

Please add test for new graph options functionality

[aMahanna]

Please add test for new graph options functionality

Unfortunately there is not much to validate here, since python-arango cannot provide any information the main graph options like shard_count, replication_factor, write_concer, etc. I.e if we did set these values, there would be now way for us to systematically validate for it (a quick manual test however shows that these work just fine)

It does allow you however to access orphan_collections via db.graph(name).properties(), but DGL has no concept of orphaned data (nodes must always be linked via edges).

As a result, adding something like arangoml/networkx-adapter@f64c8e0 is just not possible.

[aMahanna]

Setting PR to draft, as I need to redefine the scope in anticipation of arangodb/python-arango#207