SME2 baselines for simd-loop 216-221/223; matrix-loop n rounded to multiples of 256

simd-loop workload enlarge

Replace pre-PR#15 conv2d-only trace with current simd-loop set (definitions + workloads + reference/autovec solutions)

Update bench-trace: sync definitions, solutions, workloads, traces (#1)

Add simd-loop workloads (edge + perf sizes for 23 loops)

Upload folder using huggingface_hub