Improve write logic

[LeeSmet]

Right now a backend setup is generated, and if any 1 zdb fails, the entire write fails. It should at least be possible to do a "degraded write" if at least data_shards backends are online. In this case, we should probably also mark the metadata in a separate tree as "degraded" for this file, thus knowing that it later needs to be fixed (when config changes).

While we are at it, it might also be possible to further improve the backend selection. We could generate new configs in a loop, removing backends which aren't healthy as we discover them, and remember the least degraded config in case we can't find a fully healthy one. This solution should find a fully healty setup if one is available at all given the constraints

[LeeSmet]

With the development of a monitoring daemon, this issue is reduced to just pruning unhealthy backends before writing to them. If a degraded write feature is needed in the future, a new tracking issue can be opened for it