Default to parallel checkpointing algorithm, with sensible number of threads in the next release.

Checkpoint version would be an optional configuration parameter. The default would then change across releases. From the perspective of client it doesn't really matter which checkpoint it is restoring from.

The following will have to be tackled:
* E2E test that shows interoperability of two versions.
* Likely we need fallback/extend peerset in case of missing checkpoints/chunks.
* Both old and new checkpoint should be requested equally to preserve load for the initial adopters.
* We may want to make number of subtrees a consensus parameter, just like chunk size. Maybe keep it simple for start...