We could be more consistent with `Ipv6Network` prefixes

While investigating #5665, I ran into a few places where we're using Ipv6Network for types that we know should have a specific prefix size. Two examples are https://github.com/oxidecomputer/omicron/blob/9c90e4b54694e8b4bec1884306d2626dcd062246/common/src/api/internal/shared.rs#L162 and https://github.com/oxidecomputer/omicron/blob/9c90e4b54694e8b4bec1884306d2626dcd062246/nexus/db-model/src/rack.rs#L19

Both of these should always be /56 networks, but the types don't enforce that, which allowed #5665 to sneak in in the first place. I think these should be Ipv6Subnet<RACK_PREFIX> (and the same for other places where we know we should have a specific prefix size). I don't think this is a trivial fix because at least the first of those examples above is serialized in the bootstore, so any change would either need to be backwards compatible or deal with migrating the format.

Start with the Ipv6Network uses in common/src/api/internal/shared.rs around line 162 and nexus/db-model/src/rack.rs around line 19, then review #5665 to understand the prefix issue. Check how the first type is serialized in the bootstore; done means enforcing /56 prefixes where they are known while preserving or migrating the serialized format safely.