Cinder (block storage) support: create, resize, snapshot with small/medium/large profiles

[berendt]

Closes #30

First Cinder (block storage) slice for openstack-tester: generate, apply, status, report, and cleanup for volumes and snapshots across small/medium/large profiles, wired through the same auth, metrics/OTEL, run-record, and CLI plumbing that Neutron already uses. Volumes are created, optionally extended (resize), and snapshotted; cleanup tears them down in reverse dependency order by ostester:run metadata.

Change set (in commit order)

1. 14bbff8 Move the Neutron profiles to scenarios/neutron/ — a hard, bisectable move ahead of a second service: the flat scenarios/{small,medium,large}.yaml become scenarios/neutron/<name>.yaml, and every reference (embed directive, profile tests, chaos test, Makefile, README) is updated in one commit. No fallback for the old flat paths.
2. 27ebe87 Add NewBlockStorageClient to internal/config — an authenticated BlockStorageV3 client mirroring NewNetworkClient line for line, reusing the same clouds.yaml resolution.
3. 8247c2e Extract shared resource identity; add run-record service field — pure refactor plus additive fields. The Kind/Logical/Name/ID identity struct moves out of neutron into a shared internal/resource package (neutron.Kind/neutron.Resource become type aliases, so neutron files compile unchanged). run.Record gains Service (omitempty, read as neutron when absent so pre-Cinder records load unchanged) and VolumeType; neutron apply/chaos/monitor now stamp Service: "neutron" explicitly.
4. 112e179 Add the Cinder plan model — internal/cinder/plan, mirroring internal/plan: pure data, byte-identical JSON for the same input, a Plan of Volumes and Snapshots. ResizeToGiB is 0 when never resized, otherwise strictly greater than SizeGiB; snapshots reference their source volume by logical name. Validate resolves references and enforces size invariants; TotalGiB is shared by Summary and the quota pre-check.
5. 45c6331 Add the Cinder scenario schema and deterministic generator — internal/cinder/scenario with its own strict YAML schema (volume count and distributions for size, resize ratio, growth, snapshots-per-volume) under UnmarshalStrict. Generate expands scenario + seed into a byte-identical plan in a fixed draw order, guarding the resize decision behind a positive ratio (mirroring the neutron generator). A small.plan.json golden fixture locks byte-stability.
6. f38a64d Add the Cinder client: volumes, snapshots, quota pre-check — internal/cinder wrapping gophercloud v2 block-storage calls through the same timed metrics+OTEL seam as Neutron. Deterministic ostester-<id>-<logical> names and ostester:run/ostester:type metadata ride in the create request (no separate tag step). ExtendVolume adds the extend operation label; WaitForReady polls to available and fails on terminal error* status; WaitForGone backs cleanup ordering. Volumes are filtered server-side by metadata; snapshots (no metadata filter in gophercloud v2.12.0) are listed and filtered client-side. PrecheckQuota itemizes an oversized plan against global and per-type volume/snapshot/gigabytes quotas, folding Cinder's 413 into the quota classification. The three telemetry instrument descriptions and the metrics package doc drop their Neutron-only wording.
7. 6697249 Add the Cinder staged executor — internal/cinder/executor with three strictly ordered stages: create-and-poll every volume, extend only resize-targeted volumes and re-poll, then snapshot. Snapshots are grouped by source volume — sequential within a group, groups concurrent up to --concurrency — so a backend rejecting a snapshot of a still-snapshotting volume is tolerated without losing cross-volume throughput. Terminal volume errors fail the stage (still recorded for cleanup); a readiness deadline during a live run is warned and tolerated; quota errors fail fast and return the partial Created. The retry/backoff policy is a deliberate local copy of internal/executor (which hardcodes the neutron classifiers). Fake-backed tests cover ordering, extend-only, snapshot serialization with cross-volume concurrency, quota fast-fail, terminal error, transient retry, cancellation, and the concurrency bound — all under -race.
8. 4645560 Add metadata-based Cinder cleanup — executor.Cleanup tears down in reverse dependency order (snapshots, then their source volumes, since Cinder rejects deleting a volume with live snapshots). Each kind is discovered by ostester:run metadata and unioned (dedup by id) with the run record's created list. Snapshot deletion is asynchronous, so each is waited on until gone (bounded by --timeout) before volumes are deleted. A 404 counts as success (second sweep is a no-op); the first non-404 error stops with the count so far.
9. 1b6e5f4 Add the cinder CLI namespace — the cinder command as a sibling of neutron (generate/apply/status/report/cleanup), wired into root and the global flags. apply takes optional --volume-type (unset = cloud default), recorded in the run record and fed to the per-type quota pre-check; --dry-run prints the plan summary before any client exists. Shared plumbing is factored where it helps: writePlanOutput, writeStatusTable over a small observer interface, resolveRun returning the whole record, and a requireService guard (empty service treated as neutron) that stops a service's commands operating on another service's record. Command tests cover registration, generate stdout/set/seed, the cloudless dry-run, the cloud-required paths, and the cross-service guard both directions (including a legacy record read as neutron).
10. 2bc7058 Ship the Cinder small/medium/large profiles — scenarios/cinder/{small,medium,large}.yaml sized per the issue's table and embedded alongside the neutron profiles. small (5 volumes, 1–5 GiB, 0–2 snapshots) is deliberately sized to fit Cinder's common default quotas (10 volumes / 10 snapshots / 1000 GiB) so it runs against a fresh project without raising anything; medium/large need raised quotas, guarded by the pre-check. Profile tests lock parse/name/validate/generate, the golden equality and default-quota fit for small, and large strictly larger than medium.
11. 4326a75 Document the Cinder slice in the README — status paragraph and scope-by-phase table note the first Cinder slice; §6 example and built-in profiles point at the moved scenarios/neutron/ paths; §9 attribute value sets gain the new volume/snapshot kinds and extend operation plus a note that cinder apply --otel flows through the existing instruments unchanged. A new Cinder section (§15) documents commands, scenario schema and sizing, --volume-type semantics, the quota pre-check, and the snapshots-then-volumes cleanup; the layout tree grows internal/resource, internal/cinder, and the split scenarios tree; trailing sections renumbered.
12. 4590b19 Put ctx first in writeStatusTable — small follow-up: context.Context first parameter for the shared status-table helper, matching the rest of the codebase.

Notes for the reviewer

• Breaking path move (14bbff8): the old flat scenarios/{small,medium,large}.yaml paths no longer resolve. Anything pinning those paths (embedded-FS keys are now neutron/<name>.yaml) must be updated. This is intentional and called out in the commit.
• Deliberate duplication: the executor's retry/backoff policy and the report builder are copied/instantiated per service rather than abstracted, because the shared internal/executor retry policy hardcodes the neutron classifiers. This trades a small amount of duplication for keeping each service's classifiers independent.
• gophercloud limitation: snapshots have no server-side metadata filter in gophercloud v2.12.0, so snapshot discovery lists and filters client-side on the ostester:run metadata the tool itself wrote.
• No divergence from the issue's intent was found in the commits — the small/medium/large profiles follow the issue's sizing table, and small is sized to fit default Cinder quotas.

---

Implemented by planwerk-agent 6b3f3ea with Claude:claude-opus-4-8