Add ability to execute arbitrary data loading and ETL operations on graph input data

[rlratzel]

DGL users are responsible for loading and cleaning the graph input data used to create the DGL graph objects. Since loading the data needs to be done server-side, GaaS clients need to be able to execute calls to load files and clean/reorganize data for creating graph objects.

Options include:

• Client API that can tell the server to import an arbitrary module, then run functions provided by that module. Users would write the necessary ETL code as a python module that can be loaded and run.
• Client API that allows users to send UDFs over in the form of code strings which can them be eval'd server-side to read and update data.

Both of the above can be scoped specifically to graph creation in order to simplify the implementation, however, the ability to run arbitrary code could be generally useful for other things.

[rlratzel]

@BradReesWork also suggested the following in #7 :

Suggestion: One of the start-up argument should be a reference to a Python file that contains the data loading process.
The file must contain a function: def load_data(): -> PG
loader = import(sys.argv[1].replace('.py', ''))
PG = loader.load_data()