Simplify NLLB: remove 503 handling, clean cache logic, remove unused None path

Fix NLLB cache: use session_state, never cache None; remove GPT-5 temperature param

Cache NLLB translations (1h TTL); fix non-blocking semaphore acquire on server

Remove sleep debounce; 503-on-busy for NLLB; fix GPT-5 max_completion_tokens

Debounce auto-retranslate on rapid lang/model changes; fix GPT-5 max_completion_tokens

Increase NLLB request timeout to 90s to handle CUDA warmup latency

Inline translating indicator with animated dots; no spinner overlay

NLLB first; always-active button skips API when unchanged; inline translating placeholder

Live button state on keystroke via JS input listener; no blur wait

Disable translate button when text unchanged; whitespace-only edits don't count

Auto-retranslate when model or language changes after a prior translation

Update models: GPT-5+4o, Gemini 2.5+3 Flash; remove dark mode toggle

Remove Gemini 1.5 and emojis from dropdown; fix dark mode JS via components.html

Fix: use gemini-2.5-flash, add https scheme guard for TRANSLATE_API_URL

Rename M2M100→NLLB in UI; update API URL to satranslate.hausanlp.org

Fix model names: gemini-2.0-flash, gpt-4o, M2M100 AfriScience-MT; add dark mode toggle; friendly model dropdown names

first commit

Duplicate from streamlit/streamlit-template-space