Cluster-wide configuration: blobstores, value types, ...

[bruno]

Currently registering blobstores and value types is done using programmatic API calls.

Since these should be the same across all nodes (servers or clients), we should think about storing their configuration in zookeeper.

For non-standard implementations (user-provided value types or blobstores) there is also the difficulty of making the code available on all the nodes. Is this a problem of the user, or can we do something around this? At least properly handle the situation where implementations are missing.

While thinking about this, we should already think about future needs too: uniform configuration of pre-save hookds, of secondary tasks (cfr. WAL), ...

We might also consider removing the methods for registering these things from the public APIs.

[bruno]

In the short term we will address this through #69.