DAL/Node: improve catchup for attester nodes

The following discussions from !16558 (merged) should be addressed:

• @eugenz started a discussion: (+1 comment)

  I wonder why we don't just save the last processed level when done with the clean-up.

• @Saroupille started a discussion: (+2 comments)

  If the last processed level is very far, why cleaning up 8 levels from the last processed level is not enough? More than cleanin up things, when in operator mode we also need to register the skip list cells I guess? But this is true only for the operator, no?