Reject unhealthy local k3s substrate states

[waqasm86]

Summary:

• compare the advertised k3s InternalIP with current host interfaces
• parse pod phase and container readiness instead of treating kubectl success as workload health
• document the complete 64-repository and live-cluster audit
• correct the documented default repository paths

Validation:

• gofmt and module verification
• unit and race tests
• go vet and build
• govulncheck
• live validator now rejects the stale node address and unhealthy GPU Operator state
• GitHub Actions validate: success