We currently account disk image bytes per btree for shared disk image, so two btrees having shared disk image are counting the disk bytes twice, this causes a slightly overcount for 1. __evict_btree_dominating_cache when we decide if a single btree is occupying >= half of our target's cache usage 2. __evict_walk_target when we calculate how many pages to queue for a given tree. As these two functions are using the total cache as the denominator which is shared disk image bytes deduplicated value.

We should fix this by counting the total duplicate shared disk bytes, and account those bytes in the denominator when doing the calculation.