metal : unify mul_mv_id kernels (llama/6556)

ggml-metal.m

@@ -1941,7 +1941,12 @@ static enum ggml_status ggml_metal_graph_compute(
                                     {
                                         nth0 = 4;
                                         nth1 = 16;
+                                    #if QK_K == 64
+                                        pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_NL_F32].pipeline;
+                                    #else
                                         pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_MUL_MV_ID_IQ4_XS_F32].pipeline;
+                                    #endif
+
                                     } break;
                                 default:
                                     {

ggml.c

@@ -11074,7 +11074,6 @@ static void ggml_compute_forward_mul_mat_id(
         }
 
         // initialize matrix_row_counts
-        GGML_ASSERT(wdata == wdata_src1_end);
         memset(matrix_row_counts, 0, n_as*sizeof(int64_t));
 
         // group rows by src0 matrix