The current cost formula for index scans and index probes + fetch overestimates the number of random IOs that will be performed because it assumes that every index entry will trigger a random IO.

The task of this ticket is to implement a better estimation of the number of page fetches (random IOs) that an index scan (or index probe) will perform. We will use the Y_wap formula described in theorem 3 of "Index Scans Using a Finite LRU Buffer" by Mackert and Lohman (https://dl.acm.org/doi/epdf/10.1145/68012.68016).