Surah Al-Fatiha revision check (prototype) build v7b

Test prototype - not a finished app. Al-Fatiha only. First load downloads a speech model of about 36 MB (then cached), so use Wi-Fi. Word updates are near-live, not instant. Speed on real phones has not been tested yet. It can wrongly mark a correct word, and it can miss real mistakes. It checks word letters only: it cannot see vowel-mark (harakat) mistakes at all, and it does not judge tajweed or pronunciation. Known blind spots: it accepts both مالك and ملك (two valid readings) because the model writes both the same; it cannot hear long-vowel (alif) length; and a one-letter difference is shown amber, not red. It also hears other people talking nearby and may show their words as yours; it cannot tell voices apart. Do not rely on it for your Quran learning.
Privacy: speech recognition runs in your browser, and the page does not upload or save your audio. Loading it fetches the speech-runtime code from cdn.jsdelivr.net, so that service can see your IP address; the model file (36 MB) is fetched from huggingface.co (a fixed version, checked against a known SHA-256 before use), so that service can see your IP address too.