cleanup: remove the translation module — the guard is English-only now (#105)

Add the v1 eval dataset: 100 valid + 100 invalid + 20 agentic

Guard: replace the injection classifier with embedding-space membership