fix(metrics): correct storage backend label and doubled _total suffix

[sionsmith]

Summary

Fixes the two metric defects reported in strimzi-backup-operator#50:

• backend="filesystem" on S3 backups — SegmentFlusher::write hardcoded "filesystem" when recording kafka_backup_storage_write_bytes_total and kafka_backup_storage_write_latency_seconds, regardless of the configured backend. StorageBackend now exposes a backend_name() method (s3, azure, gcs, filesystem, memory) and the flusher labels metrics with the backend it actually wrote to. To be clear for the reporter: segments were always streamed directly to object storage — only the label was wrong, no PVC is needed.
• Doubled _total_total suffix — counters were registered with an explicit _total suffix, but prometheus-client appends _total to counters at encode time, so series were exposed as e.g. kafka_backup_records_total_total / kafka_backup_storage_write_bytes_total_total. Counters are now registered un-suffixed, so exposed series finally match the names documented in docs/configuration.md (kafka_backup_records_total, kafka_backup_storage_write_bytes_total, …). status-watch continues to accept the old doubled names so it can watch pre-0.15.12 endpoints.

Breaking (dashboards only): any dashboard/alert built on the doubled *_total_total names must move to the documented single-_total names.

The third symptom in that issue — static kafka_backup_lag_records / kafka_backup_lag_records_sum / kafka_backup_snapshot_records_remaining — is the cardinality-budget freeze already fixed in v0.15.11 (#131): the reporter's long-running job predates the v0.15.11 image rollout. Verified below that v0.15.11+ gauges track progress correctly.

Verification

Reproduced locally against apache/kafka:3.7.1 + MinIO with 3.2M records (~1.9GB) in a 6-partition topic, snapshot mode (stop_at_current_offsets: true), scraping /metrics every 400ms during the run:

	v0.15.10 image behaviour	this branch
kafka_backup_lag_records	frozen at initial values for entire run	counts down to 0 per partition
kafka_backup_snapshot_records_remaining	absent	3,191,285 → 0
kafka_backup_lag_records_sum	absent	3,191,285 → 0
storage write bytes	kafka_backup_storage_write_bytes_total_total{backend="filesystem"}	kafka_backup_storage_write_bytes_total{backend="s3"}

• cargo test --workspace — 259 tests pass, including new regression tests:
  • counters_expose_single_total_suffix (no _total_total anywhere in encoded output)
  • test_segment_flush_labels_actual_backend (flusher labels the real backend, never a hardcoded filesystem)
• cargo clippy --workspace --all-targets -- -D warnings clean

Follow-ups

• kafka-backup-docs reference/metrics.md documents the _total_total names and should be updated to the single-_total form once this ships.
• strimzi-backup-operator DEFAULT_BACKUP_IMAGE should be bumped to v0.15.12 after release.

🤖 Generated with Claude Code