[type safety] Strongly typed UUIDs

We use UUIDs for everything in Omicron - identifying internal objects, external objects, and probably just for fun in a couple other spots too.

These all use the uuid::Uuid structure, which is functionally sufficient, but less safe than it could be. For example, in PRs like https://github.com/oxidecomputer/omicron/pull/741 , where a "Disk ID" was replaced by a "Volume ID", the underlying types are both UUIDs, but they're fundamentally different.

Building a simple "newtype" wrapper around a UUID for each object would prevent this misuse, along the lines of:

#[derive(Clone, Copy, Serialize, Deserialize, Debug)]
struct DiskID(pub uuid::Uuid);

impl std::ops::Deref for DiskID {
  type Target = uuid::Uuid;
  fn deref(&self) -> &Self::Target { &self.0 }
}

#[derive(Clone, Copy, Serialize, Deserialize, Debug)]
struct VolumeID(pub uuid::Uuid);

impl std::ops::Deref for VolumeID {
  type Target = uuid::Uuid;
  fn deref(&self) -> &Self::Target { &self.0 }
}

These structs would be usable wherever a standard uuid::Uuid object could be used, but would avoid being interchangeable with each other.

Start by inventorying the repository's uses of uuid::Uuid and identifying which UUIDs represent distinct object types. Review the proposed newtype approach and its serialization and API implications. Done means relevant object identifiers are distinct types and cannot be passed interchangeably.