Sync Dream Mode Qwen3 local inference policy

[sheawinkler]

Summary

• Syncs Dream Mode Qwen3/local inference policy to the public lane.
• Adds patient bounded Dream Mode LLM controls and deprecated qwen2.5-coder replacement.
• Expands runtime policy guidance for SGLang, vLLM, MLX/mtplx, llama.cpp, TGI, TensorRT-LLM, and resource-aware local model sizing.

Verification

• GOCACHE=/Volumes/wd_black/codex_go_cache GOMODCACHE=/Volumes/wd_black/codex_gomod_cache go test -count=1 ./... in services/gateway-go
• bash -n scripts/inference_runtime_policy.sh scripts/inference_backend_guard.sh
• python3 -m py_compile scripts/agent/contextlattice-dream