Treat invalid IRIs as errors in conversion to RDF

[aucampia]

Currently invalid (i.e. non absolute) IRIs are ignored and tripples
containing non absolute IRIs are omitted from the RDF resulting from
to_rdf.

This patch changes this behaviour so that non absolute IRIs result in
exceptions instead.

See #137

I am not sure what to do about testing, it seems there are no general unit tests, but rather just some scaffolding to run the w3c tests (which fail).

If you want I can add some general unit tests for you, just need to agree on the framework I guess.

[gkellogg]

The spec specifically calls for such IRIs to be ignored, so raising an error would cause some tests to fail. I think the best thing to do is issue a warning, but a processor flag could cause an error to be raised.

[aucampia]

Thanks for the info @gkellogg

I think maybe to keep things clean it is best to add two options: strict and warn. If strict is set to True then raise when encountering an invalid IRI (i.e. not a valid absolute IRI), if warn is set to True then warn in the same case.

strict and warn could then be used for other similar cases in future.

Not sure what you think is best, feedback is appreciated.

[aucampia]

The other option is maybe to just have warn and errorOnInvalidIRI - but I worry that this will lead to proliferation of errorOn... options.