query :: provide decent source-composite approach (cleanup current halfway attempt)

[mpo-vliz]

historically the kg2tbl introduced a halfhearted aggregate of results over multiple URI
i.e. the GraphSourcde.build(*list_of_url) approach suggests one could be hitting multiple sources in one go

• however it failed to mix results from different kinds of sources
• never actually tested querying e.g. 2 file-dump sources
• we haven´t been actually encountering use cases for this

as a consequence some weird code constructs are still there

in recent fix of #85 I also just broke that promise for the SPARQLGraphSource case
(as it now only will query one endpoint only)

as things are now I would propose to consider re-introducing this

• only later (if we really see a need)
• through a composite pattern --> introducing a CompositeGraphSource object that holds a number of independent GraphSource objects

this will more naturally solve the current problem of not supporting a mix of sources too!

• with the recent stronger pandas usage, we could probably defer the aggregation of the results on pandas level too?

at the time we go into this, we should some time to cleanup the code to remove other historic non-sense stuff in this area (e.g. FileGraphSource can be changed to work on only one file as CompositeGraphSource will solve mixes higher up)