cache_state: afk/cache-editing/thinking latches are dead-stores; 1h prompt-cache allowlist never populated #285

File: src/state/cache_state.py:85-103

Per the in-code audit notes (M9):

• fast_mode_header_latched is the only latch with a real integration (src/utils/fast_mode.py).
• afk_mode_header_latched — dead-store; AFK toggle component not implemented.
• cache_editing_header_latched — dead-store; no Python equivalent of TS's GrowthBook cache-editing treatment.
• thinking_clear_latched — dead-store; src/utils/effort.py does not expose the thinking-flip-after-cache-miss event.
• prompt_cache_1h_allowlist defaults to empty, so 1h caching is permanently dormant (only 5m caching works). Population was left to a future GrowthBook-integration WI.

Impact: Cache-telemetry headers under-report toggle activity vs the TS reference, and 1h prompt caching can never engage regardless of model/config.

Fix sketch: Wire each latch at its source event as the corresponding feature lands, and add a config-backed (non-GrowthBook) way to populate prompt_cache_1h_allowlist so 1h caching is reachable in the Python port.