fix: track created disks in the state file so an interrupt cannot orphan them

[tas50]

The bug

Pressing Ctrl-C during a kitchen create that uses extra disks orphans a billable disk with no record of it anywhere:

=== disks ===
tk-datadisk-ubuntu-2204-fa573c-extra-disk  20GB  (no users)

=== state file ===
--- {}

=== kitchen destroy — the user's only cleanup lever ===
       Finished destroying <datadisk-ubuntu-2204> (0m0.00s)

=== disks after destroy ===
tk-datadisk-ubuntu-2204-fa573c-extra-disk  20GB   <- still there, still billing

Why

created_disk_names was @created_disk_names ||= [] — in memory only. The cleanup that consumes it hangs off create's rescue => e, which catches StandardError. Interrupt is not a StandardError, so Ctrl-C skips the rescue, and the process exits with the only record of the disk in a variable that just died with it.

Standalone disks are created before insert_instance and bill from creation, so this window is wide open — it is the whole "Creating a N GB disk / Waiting for disk to be ready" phase.

The fix

Record the disks in the state file, which Test Kitchen writes whatever happens (ensure state_file.write(state)), and have destroy delete anything still recorded once the instance is gone. This is the treatment the instance itself already gets.

Two supporting changes make that reachable:

1. Pin the zone before anything billable exists. In region mode the zone is chosen at random, so cleanup would otherwise have no idea where to look.
2. Record the server name before the insert is issued, not after it returns. Live testing caught this: on one interrupt GCE created the instance but the request never returned, so state had the disk but not the instance holding it, and cleanup failed with resourceInUseByAnotherResource. Recording the name first closes that window, and fix: clear the state file when the instance is already gone #159 makes destroy tolerate a name that never became an instance.

Verification

• bundle exec rspec — 331 examples, 0 failures. Five new specs; three of them fail
  when only lib/kitchen/driver/gce.rb is reverted to main.
• bundle exec rubocop reports 5 Lint/DuplicateMethods offenses, all of them
  already on main — fix: leave room for the disk name when generating an instance name #162 and fix: downcase instance names instead of mangling them #163 each added the same five private helpers to
  gce.rb and neither merge removed the duplicate. Not introduced here, and fixed
  separately.
• Against a real GCE project, interrupting at the same point as the original repro:

# state file after Ctrl-C — now has everything cleanup needs
zone: us-central1-a
server_name: tk-datadisk-ubuntu-2204-0a8388
created_disks:
- tk-datadisk-ubuntu-2204-0a8388-extra-disk

kitchen destroy
       GCE instance <tk-datadisk-ubuntu-2204-0a8388> destroyed.
       Unable to locate disk tk-datadisk-...-extra-disk   <- already gone with the instance

Final state: 0 instances, 0 disks.