nx_gguf_fixture_tiny.nx

nx_gguf_fixture_tiny.nx -- reusable tiny-Llama GGUF fixture builder. Encapsulates the ~150 lines of byte-by-byte synthetic GGUF construction that test smokes for nx_llm_run_v2 and the v2 LLM actor adapters currently duplicate. V1 builds the canonical n_layers=1, hidden=2, head=2, vocab=4, ffn=4 shape with all-zero F32 weights -- the same fixture nx_llm_run_v2_test.nx ships -- plus the BPE vocab (4 single-byte tokens 'a' 'b' 'c' 'd'), PRNG, model spec, and a vocab-id-to-byte map for autoregressive smokes to roll the previous token's bytes into the next prompt. One call -> entire test fixture ready to feed nx_llm_generate_one_v2 or any actor adapter that wraps it. Closes the combined-autoreg- over-real-transformer smoke gap documented in nx_actor_role_llm_v2_autoreg_test.nx. Honest scope: V1 is a SINGLE-shape fixture -- there are no knobs. Future V2 will add a builder-pattern (caller-set hidden_dim / vocab_size / n_layers). V1 keeps the API surface tiny.