Switch the MINIO scrape pipeline to Metrics V3 and configured bucket metrics

[Vonng]

Parent: #763

Problem

The public Pigsty scrape configuration still defaults to /minio/v2/metrics/cluster. That endpoint does not cover all node/drive/resource data expected by modern SILO Console widgets, while SILO standardizes on Metrics V3. Bucket metrics also need an explicit per-bucket path.

Proposed baseline

Use the simple, complete topology first:

• scrape /minio/metrics/v3 for every SILO instance;
• for every bucket declared in minio_buckets, also scrape /minio/metrics/v3/bucket/<bucket>;
• keep one logical job="minio" and preserve cls, ins, ip, instance, and flavor;
• select the per-target bucket path through file-SD/relabel metadata;
• keep existing RustFS health-probe behavior intact;
• make legacy MinIO behavior explicit instead of sending unsupported V3 paths to it accidentally.

This intentionally favors complete coverage and a small configuration surface. For N instances and B managed buckets, the expected scrape-target count is N × (1 + B). Measure the actual series count and duplicated cluster/bucket-wide series before adding a more complex designated-target optimization.

Required validation

• Inventory the V3 metric families and series produced by single-node and distributed SILO.
• Quantify per-bucket series growth and verify that only Pigsty-managed buckets are added by default.
• Identify V3 cluster-wide gauges repeated by each instance; retain the raw data but document the correct dashboard aggregation.
• Verify target uniqueness. Do not manufacture uniqueness by putting a bucket name only in the display-form instance label if another stable label is required for querying.
• Verify scrape authentication, TLS, reload, and target health.

Acceptance criteria

• Every SILO instance exposes a healthy V3 root target.
• Every configured bucket exposes a healthy bucket target; no undeclared bucket is scraped by default.
• Series retain stable job, cls, ins, and bucket identity labels.
• Target and series counts are documented for at least 1-node and multi-node test clusters.
• RustFS and any retained legacy MinIO profile do not regress.
• VictoriaMetrics reload succeeds and post-reload queries prove root, node/drive/resource, and bucket families are present.