v0.16.0

Follow-up to the issue #2 rerun on VBR 13.1.1.18, where a complete 123-backup
ranking took 21 minutes (4 s of it local CPU) and needed a 300 s timeout.

Added

• Scope for backup_storage_ranking: backups (ids or names — a backup is
  named after its job) and repository (id or name); CLI --backup (repeatable)
  and --repository. Filtered locally, since /backups has no repository
  filter. A scope matching nothing is an error rather than an empty ranking. New
  payload fields scoped, scope, backupsInScope; backupsTruncated is now
  measured against the scope, and a scoped ranking carries a caveat (CLI:
  SCOPED) because it is not an estate-wide ranking.
• Parallel reads, opt-in: concurrency (1–8, default 1; CLI
  --concurrency). Sequential by default because parallel reads are
  unmeasured on a real VBR, whose server is already the bottleneck. The payload
  is identical at any concurrency, a hard error in one read returns at once,
  and token renewal is serialised so parallel requests meeting an expired
  token log in once.
• Progress for veeam-aiops backup ranking on stderr, so --json stays
  parseable.
• Unmatched owners are looked up: a file naming an owner its own backup does
  not list is resolved once per id through /backupObjects/{id}. When the
  server knows the object (typically a machine moved to another job) its bytes
  are charged to that machine and counted in recoveredOwnerStoredBytes;
  otherwise they stay in unmatchedOwnerStoredBytes. unmatchedOwners reports
  each id's resolution (largest first, at most 50 entries, with
  unmatchedOwnersTotal / unmatchedOwnersTruncated); ownerLookupsSkipped
  counts ids past the 200-lookup budget, which get their own caveat instead of
  the namespace one.
• Scale-out repository names are matched by the repository scope and shown
  by backup_object_storage_usage.
• Timeouts are named as such in the ranking: unreadableBackups[].timedOut,
  a caveat and a CLI line pointing at the target's timeout in config.yaml.
  VeeamApiError.timed_out exposes the same fact to any caller.

Changed

• unmatchedOwnerFiles / unmatchedOwnerStoredBytes now count only owners the
  server could not resolve; resolved ones move to recoveredOwner* and into the
  machine's storedBytes. unresolved* is still ownerless + unmatched, so it
  drops against v0.15 by whatever was recovered. The chargeback criterion is
  unchanged: unmatchedOwnerFiles must be 0.

Fixed

• A backup the server lists without an id was skipped silently; it is now
  reported in unreadableBackups.
• A login that timed out was reported as an unreachable server; it now says it
  timed out and is flagged timed_out.
• An unreadable backup's error was cut at 200 characters without saying so; it
  now ends with an ellipsis.