The big disagg merge originally included a change to the block cache's heuristic when deciding whether to compress a page. This caused a performance regression for non-delta cases, since the change was pretty dumb. WT-14609 backed that change out, but we still want something a little smarter that understands the disagg case.