Implement versioning with Pijul

[multun]

Hey!

I just talked with Pierre-Étienne Meunier, the creator of Pijul. Here's what came out of our discussion:

• pijul can probably work for our use case if enough work is put into it
• that amount of work seems low enough to be realistically undertaken by us + assistance by him

Storage

Pijul's storage is two fold:

• only patches really need to be preserved. Their order does not matter, as it is a CRDT. These patches depend on other patches. These can be streamed from production systems into a backed up database
• the live pijul server relies on a special key value called sanakirja to perform its routine operations. It could be replaced by a distributed KV, but really shouldn't: it is performance critical, it has reference count for fork (for branches), and its content can be rebuilt from the patches anyway.

Structured data

Pijul is a CRDT which works on bytes. We ideally do not want to change that, as it would be fairly difficult.
An easier option would be to produce Pijul binary patches on from changes on json. That's not easy, but pretty
manageable.

A concern is that we need to make patches that keep json files valid.

Scale and complexity

The complexity of checkout of files and the number of live (not made obsolete) patches in these files.
Applying new patches depends on the number of live patches.