fix prompt for short answer

fix to ui to visible

perf: improve LFM2.5-Audio inference quality and reliability

perf: warmup generation, async LFM loading, reduce max_new_tokens

perf: add torch.compile for LFM model on HF Space (20-60% faster inference)

fix: proper Mimi decode on HF Space — only disable dynamo on Windows, explicit device handling

fix: use Mimi streaming decoder for proper speech audio (not LFM detokenizer)

fix: Play produces no sound — switch to streaming audio + system deps

fix: correct LFM audio sample rate to 48kHz and normalize volume

fix: audio decode normalization + warmup first 5 paragraphs

feat: bounded pre-gen, cancellation, disk cache pruning, GPU lock

refactor: runtime config, Q&A flow module, storage paths, and cleanup

perf: 5 latency optimizations for TTS playback

feat: enable FlashAttention-2 for all models via runtime install + SDPA fallback

feat: add LFM2.5-Audio end-to-end Q&A variant