nx_real_gguf_test.nx

nx_real_gguf_test.nx -- LIVE FIRE on a real public-license GGUF. Reads /tmp/nx_real_model.gguf (Qwen2.5-0.5B-Instruct Q4_K_M, ~491 MB, Apache 2.0 license, downloaded from HuggingFace) through the substrate's own sys_read_file + nx_gguf_parse + nx_gguf_meta + nx_gguf_load_tensor_lazy. Pass criteria: 1. File reads in via sys_read_file (no truncation) 2. Magic + version match GGUF v3 3. tensor_count and metadata_count are non-zero and within sane bounds 4. nx_gguf_parse succeeds 5. At least one tensor lookup by name works (we look up "token_embd.weight" which every llama/qwen-class model has) 6. Lazy load returns a non-null handle with zero materialization Reports timing for each phase to /tmp/nx_real_gguf.tsv.