Low-latency HLS edge delivery

Design the edge delivery layer for Low-Latency HLS (LL-HLS) for sports/event streaming reaching 50M concurrent viewers with ~3s glass-to-glass latency over standard HTTP CDNs (not WebRTC). The challenge: LL-HLS works by chunking segments into tiny partial segments and using blocking playlist reloads + preload hints, which generates a storm of tiny, frequently-changing manifest and partial-segment requests that ordinary HTTP CDN caching handles badly. You need the latency of WebRTC-ish delivery while keeping the cacheability and scale of HTTP.