[Enh] Extended attribute selection for graph aggregation

[kohlerca]

This PR adds two features to the existing provenance graph aggregations:

1. Previously, for nodes with a given label, it was not possible to use an attribute that was not present in all nodes (an exception was raised). It is now possible to specify attributes that are not shared among all nodes. If that attribute is not present, it will be assumed as None when evaluating the grouping.

2. In addition to strings with node attribute names, it is possible to pass callables that take the form callable(graph, node, data). The callable can then access the NetworkX graph object, the node id, and the full node attributes dictionary. With this enhancement, it is possible to consider transformations of attribute values or the relationships of nodes in the graph when grouping.

For example, if trying to group files (label File) using the File_path attribute captured by Alpaca, a supernode in the aggregation would be created for every different file path when using the attribute name:

aggregated = graph.aggregate({'File': ('File_path',)})

With a callable, it is possible to find other similarities to group the nodes, such as the start of the path string in File_path:

is_output_file = lambda graph, node, data: data['File_path'].startswith("/outputs/")
aggregated = graph.aggregate({'File': (is_output_file,)})