Thank you for all your wonderful work, Chris! Just curious: is it feasible to eventually support sending the page clusters to backing swap in their compressed state to further reduce I/O? It's my understanding that the clusters get decompressed before getting sent to disk, which I presume is to simplify addressing.