Revert anti-bleed prompt mitigation (ineffective)

Anti-bleed + confusion-recovery guidance when no reading passage

Switch to bilingual LoRA model (educrate-qwen3-bi); hide <think>; EN+ES support

Point Space to org-hosted model build-small-hackathon/educrate-qwen3-sft

Professional minimalist UI (English, no emojis); SFT model; tighter decoding + anti-hallucination prompt

EduCrate — Spanish Socratic tutor (Qwen3-0.6B SFT+GRPO), runs on CPU