feat(aep-43): expand workload utilization metrics spec

[baktun14]

Summary

Rewrites AEP-43 — Workload Utilization Metrics from a one-paragraph stub into a full Interface specification.

Defines:

• Metrics catalog — minimum required CPU, memory, storage, network, and GPU metrics every conformant provider exposes per lease, with normalized akash_lease_* names and mandatory labels (owner, dseq, gseq, oseq, provider, service, replica, plus GPU-specific labels).
• Provider architecture — new operator/metrics package, off-the-shelf exporters (cAdvisor, kube-state-metrics, dcgm-exporter, node_exporter), VictoriaMetrics single-node as the reference TSDB (Prometheus permitted), with 30-day retention floor, per-tenant cardinality/QPS limits, and a safe MetricsClient interface that injects (owner, dseq, gseq, oseq) label filters at the gateway proxy.
• API surface — GET /lease/{dseq}/{gseq}/{oseq}/metrics REST endpoint and matching gRPC LeaseMetricsService, with JWT auth per AEP-64 (adds new metrics scope), resolution table from 15s up to 15m by range, and features.metrics advertised in ProviderStatus for discoverability.
• Client aggregation pattern — Console Metrics tab fans out across the providers hosting a deployment, merges series, and overlays deployment-update markers; CLI subcommand provider-services lease-metrics for headless use.
• Security/operational considerations — three-layer tenant isolation, query-cost DoS controls, side-channel guarantees, JWT replay mitigations, and an explicit non-goal for v1 around on-chain metric attestation.
• Phased migration — additive, no SDL or on-chain change required. Behind a feature flag until a 4-week soak with opted-in providers.

Preserves original authors (Anil Murty, Artur Troian) and adds @baktun14 as primary author. Updates index.json via scripts/index.js.

Why

Tenants on Akash currently have no first-class way to observe realized utilization of their leases (CPU, RAM, storage, network, GPU/VRAM), which blocks right-sizing, peak-load awareness, diagnostic baselines, and cross-provider comparison. This AEP fills that gap and is a prerequisite for downstream work on autoscaling, provider SLAs/quality scoring, and the Provider Console capacity dashboards mentioned in AEP-32.

Test plan

• CI passes (markdown/YAML/index validation).
• node scripts/index.js produces a clean diff with the updated AEP-43 entry (description, updated author list, new discussions-to, updated date).
• Render preview of spec/aep-43/README.md on GitHub for table/diagram readability.
• Cross-link references to AEP-32, AEP-64, AEP-65 resolve correctly.