refactor: delegate whatsapp verification to hugging face API to prevent render OOM

fix: move heuristic length check outside API exception block

fix: improve zero-shot robustness and add length heuristic

chore: migrate backend to hugging face space and add gh actions deployment

refactor: load AI models directly into memory for unified HF Space backend

fix: parse Gradio v5 SSE stream for HF Worker prediction

fix: switch HF Space worker from Docker/FastAPI to Gradio SDK (free tier)

feat: add dedicated HF Space worker for DistilBERT inference and rewire backend to call it

fix: prevent loading 800MB DistilBERT model into RAM to avoid OOM crash on Render free tier

fix: cast numpy string to python string to prevent FastAPI JSON serialization errors

fix: pass local model to detect_fake_news in step 4 to prevent 30s api fallback

fix: drastically reduce external API timeouts to prevent 60s proxy timeout on Hugging Face Spaces

fix: Re-enable local PyTorch model load for Hugging Face Spaces to bypass HF API DNS errors

fix: exclusively use user's uploaded model instead of fallback models

fix: add mrm8488/bert-tiny-finetuned-fake-news-detection to API fallback list for 100% uptime

feat: target vinitsingare/distilbert_fake_news with API fallback

feat: add model uploader script and set API URL to vinitsingare/distilbert_fake_news

fix: add candidate model fallback list and rich template generator for XAI explanations

fix: correct response parsing for zero-shot and standard input format for Qwen HF API

fix: migrate from deprecated api-inference to router.huggingface.co/hf-inference to fix global DNS outage

fix: migrate from deprecated api-inference.huggingface.co to router.huggingface.co/hf-inference to fix global DNS outage

fix: remove local model check from url_verifier so API can run

fix: fully remove PyTorch DistilBERT loading from startup cache to fix OOM crash

fix: resolve indentation error in fake_news.py

fix: replace local DistilBERT loading with HF Inference API for Ash-211/Fake_news_model

fix: replace local Qwen and SigLIP models with HuggingFace API to resolve RAM limits

Merge branch 'main' of https://github.com/Ash-211/ai_news_monitor

fix: replace local mDeBERTa model with HuggingFace Inference API to reduce RAM usage

feat: integrate deepfake analysis and update XAI to use Qwen locally

feat: WhatsApp bot improvements - interactive UI, pagination & dashboard landing page

Add WhatsApp bot integration with Meta Cloud API