feat: no state hashing during catch-up (state manager)

[mraszyk]

This PR optimizes the state manager to not hash the state at tip while catching up if consensus could deliver a certification for the tip height.

In more detail, the PR performs the following changes:

• introduces a new field states.certifications storing certifications delivered optimistically by consensus for future heights (to be used when catching up);
• take_tip uses the state hash from a certification in states.certifications if a certification at the tip height is available;
• commit_and_certify reuses the certification from states.certification if a certification at the committed height is available;
• remove_inmemory_states_below and remove_states_below prunes states.certifications at heights less than latest_state_height and latest_subnet_certified_height (this way, we have at most MAX_CONSECUTIVE_ROUNDS_WITHOUT_STATE_CLONING + MAX_FUTURE_HEIGHTS_TO_CERTIFY = 30 plus the certification lag - the difference between the latest resident and the latest certified height - many certifications stored in states.certifications at any time);
• a new method list_state_heights_to_certify is used by consensus to determine heights for which certifications should be delivered: [tip_height..min(tip_height + MAX_FUTURE_HEIGHTS_TO_CERTIFY, latest_subnet_certified_height)] \ {heights for which we already have certifications}.

The new functionality of the state manager is not yet invoked by consensus and thus the functionality is not yet effectively enabled by this PR.