nx_gguf_load_q4k_test.nx

nx_gguf_load_q4k_test.nx -- focused smoke for Q4_K dequant. Builds a synthetic GGUF with one Q4_K tensor (1 super-block, 256 values) and verifies the dequant: d = 1.0 (f16 0x3C00) -> d_q10 = 1024 dmin = 0.0 -> dmin_q10 = 0 scales[0..7] = {1, 1, 1, 1, 2, 0, 0, 0} (only sc[4]=2 in high half) mins[0..7] = {0, 0, 0, 0, 0, 0, 0, 0} nibble byte 0: 0x32 -> low nibble 2, high nibble 3 nibble byte 64: 0x32 -> low nibble 2, high nibble 3 all other nibbles = 0 ggml layout (32-byte groups; group g -> sub-block 2g low / 2g+1 high) + Q24 super-scale (d=1.0 -> d_q24 = 2^24 = 16777216). byte 0 is group 0 l=0 (low -> sub-block 0 out[0]; high -> sub-block 1 out[32]); byte 64 is group 2 l=0 (low -> sub-block 4 out[128]; high -> sub-block 5, sc[5]=0, out[160]=0). Expected Q24 outputs: out[0] = 16777216 * sc[0]=1 * 2 - 0 = 33554432 out[32] = 16777216 * sc[1]=1 * 3 - 0 = 50331648 out[128] = 16777216 * sc[4]=2 * 2 - 0 = 67108864 everywhere else: 0 Tests both the low-half (is<4) and high-half (is>=4) scale unpacking paths in nx_gguf_dequant_q4_k.