Try multiple CORS proxies in sequence for reliability

Keyless: call classic HF Inference API via CORS proxy (conversational task)

Use chatCompletion (conversational task) instead of removed conversational()

Use conversational task (HF routed zephyr to provider rejecting text-generation)

Fix CORS: use official @huggingface/inference browser client

Shorten short_description to satisfy HF limit

Add DocuTalk in-browser PDF chat (static Space)

initial commit