feat(logging): add LogBucket

[naxty]

Adds GCP::Logging::LogBucket — the last Cloud Logging type with a create and a
delete that the plugin did not have.

Branched off main, independent of the other batches.

Case	Result
logging-log-bucket	crud 8/8, discovery 4/4
logging-log-view (pre-existing)	passes — regression check, see below

Verified through debug-conformance.yml; the local key is 403 on this project's
Logging writes.

Why it matters

Metrics, sinks, exclusions, views, saved queries and log scopes were all
modelled — but not the buckets that views are windows onto and sinks route into.
A bucket is what decides how long logs live and where.

Notes for review

A deleted log bucket does not go away. It enters DELETE_REQUESTED and
stays for seven days so it can be undeleted, and a get answers 200 with that
state rather than 404 — so the generic Read would report a deleted bucket as
present for a week, an out-of-band delete would never leave inventory, and
discovery would keep offering buckets on their way out. Read reports such a
bucket as gone.

That is the same shape as the Cloud SQL backup tombstone fixed earlier in this
series. This time it was visible in the discovery document's lifecycleState
field and handled before the first run rather than after a failed one.

logging-log-view is in the CI run above because views are bucket-scoped and
share the path builder; it passes unchanged.

locked is modelled but deliberately unset in the fixture. A locked bucket
can never be deleted and locking cannot be undone — setting it in a conformance
fixture would strand a resource in the shared project permanently. The schema
says so where someone declaring one will read it.

What I could not verify

• Every project has _Default and _Required buckets created by GCP, so
  discovery reports two per project that nobody declared. That is the API's
  shape rather than something to filter, but it does mean unmanaged log buckets
  will appear after this lands.
• Link (Log Analytics → BigQuery) is not included: it needs a bucket with
  analytics enabled plus a BigQuery dataset, and link creation is asynchronous
  and slow. It is the one remaining Logging type.
• cmekSettings, indexConfigs and restrictedFields are not modelled; each
  needs its own verification pass.

---

Correction (2026-08-28): this body reported CRUD as 8/8. That was wrong. The
Replace step reports [~] — skipped, not passed: the harness only runs it when a
testdata/<case>-replace.pkl fixture exists, and no case in this repo has one (0 of
118). Read every 8/8 above as 7 passed, Replace not exercised. The same applies
to every GCP resource already on main — the Replace path has never been covered
here. Nothing in this PR regressed.