Define production replay scope and make Redis the documented multi-replica path

[OkeyAmy]

Problem

Nonce replay protection is correct only within the storage scope that remembers the nonce. The in-memory store is fine for development and single-instance pilots, but it is not a system-wide replay guarantee in a horizontally scaled deployment.

What to do

Make replay scope explicit in code, docs, and deployment examples:

• document memory store as single-process only
• document Redis as the required multi-replica production path
• add an operator warning when production-like config uses memory replay storage
• ensure examples that claim pilot/production posture use Redis

Acceptance criteria

• docs clearly define replay scope for memory vs Redis
• Docker Compose/operator examples show Redis for scaled deployments
• tests prove Redis rejects the same JTI across separate store clients
• verifier startup logs make the selected replay backend visible

Out of scope

• global consensus or blockchain replay tracking
• billing/spend counters