Local proxy that hard-caps AI coding-agent token spend per project. Point your agent's LLM base-URL at localhost; spendguard forwards to your real upstream, enforces a per-project budget, and returns an over-budget error instead of forwarding when the lim