feat: make trust score persistence the default — crash-resilient trust state

[imran-siddique]

Problem

AgentMesh trust scores are in-memory by default (Python dict in TrustBridge.peers). If an agent process restarts — deliberately or due to a crash — it re-enters at neutral trust, wiping earned reputation or accumulated violations. A misbehaving agent can effectively reset its trust score by crashing.

What Exists

• RedisTrustStore in storage/redis_backend.py provides persistent storage
• MemoryProvider in storage/memory_provider.py for dev/testing
• But the default is in-memory with no persistence

Proposed Solution

1. Add a FileTrustStore backend that persists to JSON/SQLite — zero external dependencies
2. Make TrustBridge accept a configurable storage backend (default: file-based)
3. On startup, load previous trust scores; on crash recovery, restore last known state
4. Add a trust_persistence config key to YAML policies
5. Document Redis backend as the production recommendation

Acceptance Criteria

• FileTrustStore backend persists scores to disk
• TrustBridge defaults to persistent storage (file-backed)
• Scores survive process restart without Redis
• Crash recovery loads last-known scores
• Unit tests for persistence lifecycle

/cc @imran-siddique