document_id missing in FARMReader predict return statement

[Timoeller]

The error is coming from FARM inferencer. The document_id is missing in the returned json.

Currently this breaks our evaluation.
Tutorial5_Evaluation.py needs a document_id coming back from FARM.
Haystack eval first checks weather the document_id is corresponding to the gold truth, then it checks if the answer string is correct. Since the document_id coming back from FARM is None it never enters the answer string check.

---

input to the FARM inferencer is a dict:

cur = {
                "text": doc.text,
                "questions": [question],
                "document_id": doc.id
            }

output of farm inferencer has set document_id in all cases to None.