Storage node sync should use checkpoints

[kostko]

After #1920 which adds a runtime-specific pruning policy which supports querying which checkpoints should be available, the storage node sync should use checkpoints when it needs to sync too far back.

Storage checkpointing also needs to be fixed to use explicit chunks where each chunk is independently verifiable to be part of the given root. Changing this a bit will also enable us to easily use our MKVS with Tendermint's state sync (assuming we also do #1898). (This was done in #2659.)