speed up setting docs by traversal paths

[alaeddine-13]

The current implementation of setting docs by traversal paths is not efficient because we have to retrieve the root documents from the storage backend.
A better implementation would be to traverse the documents from the storage backend and return the root document each time with the child. That way, the child should be referenced already by the parent and we just need to modify the child and persist the parent:
new non-debugged code:

def _gen_children_parent_pairs(docs_gen, parent):
    for children in docs_gen:
        yield children, parent

def _gen_children_from_pairs(pair_gen):
    if isinstance(pair_gen, Generator):
        for children, _ in pair_gen:
            yield _gen_children_from_pairs(children)
    elif isinstance(pair_gen, tuple):
        children, _ = pair_gen
        yield children

    @staticmethod
    def _traverse(
        docs: 'T',
        path: str,
        filter_fn: Optional[Callable[['Document'], bool]] = None,
        parent: Optional['Document'] = None
    ):
        path = re.sub(r'\s+', '', path)
        if path:
            cur_loc, cur_slice, _left = _parse_path_string(path)
            if cur_loc == 'r':
                yield from _gen_children_parent_pairs(TraverseMixin._traverse(
                    docs[cur_slice], _left, filter_fn=filter_fn,
                ), parent)
            elif cur_loc == 'm':
                for d in docs:
                    yield from _gen_children_parent_pairs(TraverseMixin._traverse(
                        d.matches[cur_slice], _left, filter_fn=filter_fn, parent=d
                    ), parent)
            elif cur_loc == 'c':
                for d in docs:
                    yield from _gen_children_parent_pairs(TraverseMixin._traverse(
                        d.chunks[cur_slice], _left, filter_fn=filter_fn, parent=d
                    ), parent)
            else:
                raise ValueError(
                    f'`path`:{path} is invalid, please refer to https://docarray.jina.ai/fundamentals/documentarray/access-elements/#index-by-nested-structure'
                )
        elif filter_fn is None:
            yield docs, parent
        else:
            from .. import DocumentArray

            yield DocumentArray(list(filter(filter_fn, docs))), parent

[JoanFM]

Closing until new community or team member raises