Qwen3.8-27B Atlassian model v0.5, now with GGUF

v0.5 of the Qwen3.8-27B we fine-tune for Forge, Jira, Confluence and JSM is out, and it beat v0.4 on every gate we measure it against, no waiver needed this time. GGUF versions for llama.cpp and Ollam