Consolidate Redis operations with retry logic and pipeline support

[mikevitelli]

Context

Redis operations are scattered across lib/redis.ts, lib/deviceToken.ts, lib/deviceCode.ts, and lib/deviceStatus.ts. Each module calls Redis directly with no retry logic, no error recovery, and no transaction support.

Issues:

• getUserSettings extends TTL on every read but doesn't handle expiry failure
• setUserSettings in deviceToken.ts partially updates settings (risk of losing fields)
• No connection error handling (Upstash outage = 500 errors everywhere)
• TTL values (90 days) duplicated across files

Proposed approach

1. Create lib/redis-ops.ts with typed wrappers for common patterns:
   • getWithTTLRefresh<T>(key, ttl, schema) - get + validate + refresh TTL
   • setWithTTL<T>(key, value, ttl) - set + TTL in pipeline
   • atomicUpdate<T>(key, updater) - read-modify-write in pipeline
2. Add retry logic for transient Upstash errors (1 retry with backoff)
3. Centralize TTL constants
4. Use spread operator for partial settings updates to preserve all fields

Acceptance criteria

• All Redis operations go through typed wrappers
• Transient errors retried once before failing
• Settings updates preserve all existing fields
• TTL values defined in one place