feat(platform): rehearse OTel scrape failure recovery

[zhouning]

Summary

• add a bounded M2c-3 runner that faults only the Gravitino Collector scrape address
• prove baseline, isolated failure detection, checked-in configuration recovery, and complete cleanup
• commit allowlisted live evidence, ADR-043, runtime inventory registration, CI coverage, and roadmap truth updates

Live verification

• baseline: OpenMetadata/Gravitino up=1 with 417/5 samples
• fault: OpenMetadata stays up=1 with 417 samples; Gravitino becomes up=0 with 0 samples
• recovery: both return to up=1 with 417/5 samples
• 3 port-forwards stopped; all 11 temporary resources removed; provider and ConfigMap identities preserved
• evidence fingerprint: c70211268b62ba2e4b78c2e6d356878d875216a4f5dd7a6b4894bbdff6460a8c

Verification

• 520 required platform tests passed
• 10 M2c-3 focused tests passed
• platform, provider metrics, base OTel, and M2c-3 validators passed
• frontend production build passed

Boundary

This proves only local ephemeral scrape failure detection and recovery. Production OTel, persistent storage, TLS, tenant isolation, alert delivery, SLO, OIDC, upgrade, NetworkPolicy enforcement, GDA writes, and production readiness remain false.

Stacked on PR #10; do not merge independently.