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.