Okay, so the approach here is to first MATCH to all :day nodes that have less than 30 :RESERVATION relationships.
Next per :day node you can MATCH out :RESERVATION relationships to :seat nodes and collect those, this will be our list of seats to exclude from the next match.
Next MATCH out :AVAILABILITY relationships to :seat nodes that aren't in the excluded list, and collect the relationships. You can take the slice of any beyond 30, since that's the segment we want to delete. (If you want this to be random you can use APOC Procedures for this to shuffle the list before taking the segment beyond 30).
Then we delete all relationships in that list. The remaining relationships will be cut to 30.
We can use pattern comprehension to do the equivalent of MATCHing and collecting.
Here's the query:
MATCH (d:day)
WHERE size((d)-[:RESERVATION]->()) < 30 AND size((d)<-[:AVAILABILITY]-()) > 30
WITH d, [(d)-[:RESERVATION]->(seat) | seat] as takenSeats
WITH d, [(d)<-[r:AVAILABILITY]-(seat) WHERE NOT seat IN takenSeats | r][30..] as toDelete
FOREACH (rel in toDelete | DELETE rel)
