Premium compact export

GGUF Q4_K_M without identity confusion.

Merge the trained adapter with its recorded base model, convert through pinned llama.cpp tooling, validate the GGUF signature, and download under your chosen model name.