Let SquadData support data from Annotation Tool

[brandenchan]

Proposed changes:
Data from our annotation has no title field but has a document_id field. This PR gives the SquadData object the ability to support this kind of data

[julian-risch]

Looks good to me so far. One concern that I have is about documents sometimes receiving the same, empty id. That could become confusing especially because later on, within DocumentStores, we handle documents as duplicates if they have the same id.

[julian-risch]

That would result in documents without a document_id in paragraph receiving the same, empty document_id. Is that really what we would like to achieve or could that later on lead to problems? What about using the document title to generate an id as a hash of the title, similar to what we do here:

haystack/haystack/schema.py

Line 117 in ac5617e

def _get_id(self, id_hash_keys: Optional[List[str]] = None):

[julian-risch]

Ah, and please don't forget to add labels to the PR.

[brandenchan]

Yes good idea, I have tried to generate document_id by hashing the actual text content since sometimes we don't have a title (e.g. data from annotation tool). Labels have also been added now!

[ZanSara]

I agree with @julian-risch about the default empty ID causing problems. For the rest, only a small style comment.

[julian-risch]

LGTM 👍

[julian-risch]

LGTM 👍