Update to 2.0.17: clamp work-group size to per-kernel device max; fix DCE-inflated throughput in CUDA WMMA/MMA and Metal simdgroup_matrix benchmarks; native mma.sync tensor-core path with full-rate FP16; joint_matrix on Intel Arc A770. OBS-URL: https://build.opensuse.org/request/show/1364433 OBS-URL: https://build.opensuse.org/package/show/openSUSE:Factory/clpeak?expand=0&rev=14