fix default model to 'lightning' for InferencePort API compatibility

update docs: USE_PERSISTENT_STORAGE env var, fix README

Initial commit - Forge: self-improving AI assistant with filesystem access for self-editing