support multi-machine in single configuration

When we're designing the serialization system (see #3) we should provide a way to index settings by machine name, or similar. A preliminary design that comes to mind is something like:

```
"benMachine1": { /* per-machine settings, existing schema */ },
"benMachine2": { /* ... */},
...
```

Start with issue #3, which describes the serialization system, and inspect the existing configuration schema referenced here. Resolve the multi-machine configuration shape and verify how existing single-machine settings should remain supported; the issue does not name files or tests to run.

Written by the indexing model from the issue text.