docs: OPERATOR-RUNTIME reflects llama-turboquant-runtime + Start-Llama-Server.cmd

docs: operator runtime vs fork binary; example models.ini preset

docs: MAINTAINING-FORK playbook for upstream sync and canonical build/

docs: how llama-server relates to this fork (build, naming, packaging)

docs: center fork on TurboQuant as project purpose; trim secondary Qwen section

docs: clarify TurboQuant as primary in-tree implementation; fix arXiv id

fix: MSVC M_PI in ggml-turbo-quant.c (_USE_MATH_DEFINES)

merge: origin/feature/turboquant-kv-cache — restore TurboQuant CUDA KV + FA (resolve BF16 vs turbo conflicts)

docs: clarify TurboQuant naming vs implemented changes on master

docs: fork differences vs upstream and README pointer

merge: sync ggml-org/llama.cpp upstream/master

turboquant: snapshot before syncing upstream llama.cpp (2026-04-03)

HIP: build eatch ci build test for a different architecture (#21337)

fix: add openssl to nix dependencies (#21353) (#21355)

ggml-zendnn : add MUL_MAT_ID op support for MoE models (#21315)