Support append / prepend APIs

Heavy append workload can put pressure on the disks

Instead of writing the whole doc, we should be able to write just the append chunky and then read across all the chunky to get the document, of course the compactor would have to do the same.