llama.cpp Release b10144 Fixes Stream Routes for Slashed Model Names

llama.cpp release b10144 refactors resumable stream routes to use a query string (`conv_id`) instead of path parameters, resolving an issue where model names containing slashes caused stream routes to fail. It also improves session recovery and loading progress visibility in the UI during model loading. This release ensures robust handling of model names from repositories that naturally contain slashes, preventing server crashes or session mismatches during local LLM deployment. It also enhances the user experience by gracefully handling page reloads and request cancellations during heavy model loading phases. The update moves the conversation ID to the `conv_id` query string where slashes survive URL encoding, and introduces a 503 status response to allow clients to poll and reconnect after page refreshes. Additionally, pending requests are now registered before the blocking autoload wait to allow proper cancellation.