fix: invalidate stale game assets after deploy

fix: prioritize typed answers for model grading

fix: fall back when preloaded models are missing

feat: add dual-model inference runtime

Revert "fix: harden generated lesson quality"

fix: harden generated lesson quality

feat: polish Secret Student game experience

fix: preserve authentication in Space iframe

feat: add embedded llama.cpp and Modal LLM runtimes

feat: polish Secret Student game flow and LLM reliability

initial commit