When a checkpoint is performed, any checkpoints being deleted have to have their allocation lists merged. Change the checkpoint "prep" code to pass each checkpoint's "block-allocated" list into the block manager as part of the checkpoint. When the checkpoint returns, the block-allocated list for the checkpoints immediately after any deleted checkpoints will have their block-allocated lists updated to reflect the changes, and the updated block-allocated lists will need to be written into the metadata file.
It should be possible to do this without holding the live block lock in the block manager, and that might be worth considering.

(See note #1 in the comments of the epic)