[flat index] Switch from objects to `vectors for rescoring BQ

[trengrj]

The flat index currently rescores using the objects bucket. This should be switched to use the decidated vectors bucket for performance (as no object metadata needs to be read).

weaviate/adapters/repos/db/vector/hnsw/search.go

Lines 558 to 570 in bf46c3d

	if h.shouldRescore() {
	ids := make([]uint64, res.Len())
	i := len(ids) - 1
	for res.Len() > 0 {
	res := res.Pop()
	ids[i] = res.ID
	i--
	}
	res.Reset()
	for _, id := range ids {
	dist, _, _ := h.distanceFromBytesToFloatNode(compressorDistancer, id)
	res.Insert(id, dist)
	if res.Len() > ef {

Reading occurs using the standard TempVectorForIDThunk which points to the object bucket. This needs to change to use the vectors bucket.

weaviate/adapters/repos/db/vector/hnsw/search.go

Lines 425 to 460 in bf46c3d

	vec, err := h.TempVectorForIDThunk(context.Background(), nodeID, slice)
	if err != nil {
	var e storobj.ErrNotFound
	if errors.As(err, &e) {
	h.handleDeletedNode(e.DocID)
	return 0, false, nil
	} else {
	// not a typed error, we can recover from, return with err
	return 0, false, errors.Wrapf(err, "get vector of docID %d", nodeID)
	}
	}
	vec = h.normalizeVec(vec)
	return concreteDistancer.DistanceToFloat(vec)
	}
	func (h *hnsw) distanceToFloatNode(distancer distancer.Distancer,
	nodeID uint64,
	) (float32, bool, error) {
	candidateVec, err := h.vectorForID(context.Background(), nodeID)
	if err != nil {
	return 0, false, err
	}
	dist, _, err := distancer.Distance(candidateVec)
	if err != nil {
	return 0, false, errors.Wrap(err, "calculate distance between candidate and query")
	}
	return dist, true, nil
	}
	// the underlying object seems to have been deleted, to recover from
	// this situation let's add a tombstone to the deleted object, so it
	// will be cleaned up and skip this candidate in the current search
	func (h *hnsw) handleDeletedNode(docID uint64) {
	if h.hasTombstone(docID) {