fix: sanitize off-target NPC speech

[JOY (JOY)]

Summary

• sanitize model-selected say intents that reference a non-nearby target by clearing target_id
• keep attack/interact target validation strict
• add regression coverage so off-target speech remains model-backed instead of degrading to fallback

Local review

• Local code-review fallback: approved, no blockers found in the scoped speech-sanitization diff
• Boundary check: this only affects dialogue target metadata; game-state mutations remain server-validated and strict

Verification

• npm run build (backend/nakama)
• npm test (backend/nakama)
• dotnet build Unity/SecondSpawn.AI.csproj --no-restore
• dotnet build Unity/SecondSpawn.Networking.csproj --no-restore
• git diff --check
• docker restart second-spawn-nakama and /v2/rpc/secondspawn_health smoke