When normalizing learning sequences, skip inaccessible sequences and also skip sections with only inaccessible sequences. This both imitates the legacy course block behavior and also avoids a failure when merging course block data with LS data when they disagree about which sequences exist.