feat(health): add /health/dependencies with onchain_rpc_ms latency check

[Preciousuche]

Closes #287

Summary

Adds a new GET /api/v1/health/dependencies endpoint that measures on-chain RPC latency and reports it as onchain_rpc_ms.

• If the call exceeds the 5s threshold, overall status is degraded (still HTTP 200 — the service is alive, just slow, so monitoring can see the number).
• If the call fails, status is down (HTTP 503).
• The timing logic mirrors the existing checkOnchainContract() pattern in HealthService.

Note on getAdmin(): the issue references a getAdmin() RPC call, but OnchainAdapter doesn't currently expose one. I used the existing read-only getContractMetadata() call to measure latency and left a comment on the issue asking whether a dedicated getAdmin() method is preferred. Swapping it is a one-line change if so.

Testing

• pnpm build (backend) passes — clean type-check.
• pnpm test:e2e health passes:

PASS  test/health.e2e-spec.ts
  Health dependencies (e2e)
    √ /health/dependencies (GET) reports onchain_rpc_ms
Tests:       1 passed, 1 total

The e2e test overrides the on-chain adapter with a fast fake, so it verifies the endpoint's behaviour without touching a live network.

Checklist

• Tests added for new behavior
• Build passes locally
• Follows existing code patterns

[Preciousuche]

Closing in favor of #357, which is a cleaned-up version with unrelated files removed.