nx_f32_llm_read_dims_test.nx

nx_f32_llm_read_dims_test.nx -- smoke for nx_f32_llm_read_dims.nx. Builds a synthetic GGUF with: general.architecture STR "qwen2" qwen2.block_count U32 2 qwen2.embedding_length U32 8 qwen2.attention.head_count U32 4 qwen2.attention.head_count_kv U32 2 (GQA: 2 KV heads, 4 Q heads -> group=2) qwen2.feed_forward_length U32 16 (no head_count_kv variant tested below too) + 1 token_embd.weight tensor with dim_0=12 (vocab_size=12) Verifies all dims are read correctly into NxF32LlamaModel.