[DNM, WIP] feat: rework BD naming to avoid requiring WWNs

[tserong]

Problem:
We can't really cope properly with disks that don't have WWNs, or whose WWNs change (the latter should never happen, but it did at least once in the past when a kernel update resulted in IDs for a certain make and model of NVMe changing).

Solution:

• Identify disks by, in order:
  1. UUID if present (will be there for provisioned LHv1 volumes and LVM volumes)
  2. WWN if present
  3. Vendor+Model+SerialNumber+BusPath if none of the above exist.
• BD CR names become non-deterministic (they're UUIDs for any new BDs after this change), so when scanning we don't check existing BDs by name, we check if there's one with the same UUID, then WWN, then Vendor+Model+Serial+BusPath.
• This automatically prevents adding devices with duplicate FS/LVM UUIDs or WWNs
• It would be nice if we could also interrogate provisioned LHv2 devices when scanned but that's rather more problematic (long story, I'll write up the details soon, TL;DR: this new implementation shouldn't be any more broken than the old one was for LHv2 disks without WWNs)

TODO: explain WTF I'm doing here in more detail ;-)

Related Issue:
harvester/harvester#6261

Test plan:
TBD

NOTE:
I'm still working on this. Don't worry too much about reviewing it yet.