The new checkpoint is the current `last_allowed_fork_level` of the new
head.
When updating the checkpoint the shell tags as invalid all blocks with
a level strictly higher to the new checkpoint that are inconstant with
it. And it removes from the disk all the block with a level lower or
equal to the new checkpoint that do not belongs to the current
chain. Though, the shell removes nothing from the disk when the
current head is below the current checkpoint: this will allow to
configure an expected checkpoint when bootstraping a node.
The first patch is very conservative and only detects new incompatible
blocks when they are stored on disk (i.e. after the validation).
Fiture patches try to detect earlier such incompatible block.
