feat(hygiene): stale flag detection and flag expiry

[trinhvandat]

Context (P2 — product)

Flag debt is the top operational complaint with flag platforms. Nothing tracks whether a flag is still evaluated or overdue for cleanup.

Scope

• Track last_evaluated_at per flag+environment (piggyback on evaluation path, throttled write — at most once per N minutes per flag/env to avoid write amplification; coordinate with caching feat(perf): evaluation caching (Caffeine) + ETag/304 on SDK endpoint #30)
• Optional expires_at on FeatureFlag; expired flags surfaced (not auto-disabled in v1)
• Report endpoint: flags not evaluated in N days / past expiry, per project
• Liquibase changeset for new columns

Acceptance criteria

• Evaluation updates last_evaluated_at without measurable latency impact (throttled/async)
• Report endpoint returns stale + expired flags with correct filtering
• Flag key immutability and SDK response shape unchanged