fix(gms): verify persistent KV installation at startup

[hutm]

Summary

Verify at worker startup that the requested persistent-KV integration was actually installed for vLLM or SGLang. Incomplete or missing hooks now fail closed instead of silently serving with ordinary process-owned KV.

Why this is needed

Configuration alone is not evidence that cache allocation and recovery hooks are active. A silent partial install would make the endpoint look healthy while providing no crash-surviving KV guarantee.

Validation

• The cumulative publication tip is byte-identical to the locally validated implementation.
• Cumulative focused validation: 994 passed, 1 skipped; the later vLLM-focused rerun passed 244 tests.
• Local TP2 crash tests passed exact-output recovery for vLLM and SGLang; final performance qualification remains tracked in DEP: Production hardening for persistent GMS KV failover #15035.
• This draft checkpoint still requires normal CI and reviewer validation before merge.

Stack

• Part 1 of 15 in the production-hardening train.
• Draft DEP and complete review order: DEP: Production hardening for persistent GMS KV failover #15035.
• Base: hutm/gms-common-16-v1-persistent-pool.

This PR is intentionally draft and stacked on its immediate predecessor.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.