DocumentSplitter updates Document's meta data after initializing the Document

[julian-risch]

Describe the bug
_create_docs_from_splits of the DocumentSplitter initializes a new document and then changes its meta data afterward. This means that the document's ID is created without taking into account the additional meta data. Documents that have the same content and only differ in page number will receive the same Document ID and thus might be unwittingly treated as duplicates in a later stage of the pipeline.

Instead of the current

meta = deepcopy(meta)
doc = Document(content=txt, meta=meta)
doc.meta["page_number"] = splits_pages[i]
doc.meta["split_id"] = i
doc.meta["split_idx_start"] = split_idx
documents.append(doc)

we should change the code to

meta = deepcopy(meta)
meta["page_number"] = splits_pages[i]
meta["split_id"] = i
meta["split_idx_start"] = split_idx
doc = Document(content=txt, meta=meta)
documents.append(doc)