Better metadata reliability

[LeeSmet]

The metadata is currently not really handled well in case of node failure. There are certain improvements to be made to make this more reliable:

The following things are needed, these are similar to regular data backends:

• Monitor metadata backends (Monitor metadata backends #61)
• Add ability to replace metadata backend in case its almost full (Reserve bigger metadata backend when an existing node is full #62)
• Add ability to reserve new metadata backend if one of the backends dies (A new metadata backend must be reserved if one is unreachable #63)

With the above, we can then implement metadata rebuilding:

• Support rebuilding from a (partial) metadata cluster to a full metadata cluster (Metadata cluster rebuilding #59)
• Support live migration of metadata cluster. We can stop serving requests during this time if needed, reliability > uptime (Swap out metadata clusters while running #60)