Remove broken Gemini explicit caching

[iamwavecut]

Summary

• remove explicit Gemini cache creation/list/use paths
• send one ordinary generateContent request and rely on automatic implicit caching
• retain pinned model, safety settings, strict output handling, and usage metadata

Evidence

• production: 33 explicit-cache preparation failures for 33 successful classifications in the sampled uptime window
• RED: wire regression observed 2 requests instead of 1
• GREEN: go test ./internal/adapters/llm/gemini ./internal/handlers/moderation ./internal/config -count=1
• focused go vet passed

Full repository verification runs in CI to avoid the previously reported local CPU saturation.