TOOLS/PERF: Initial support for DPU daemons

[brminich]

What

Initial support for offloading communication to DPU using exported memh feature

How ?

The way to start a benchmark is:

• Run daemons on client and server (can be on the same DPU or on a different ones): UCX_TLS=rc UCX_MAX_RNDV_LANES=1 UCX_RNDV_THRESH=0 ./src/tools/perf/ucx_perftest_daemon -p 1339 UCX_TLS=rc UCX_MAX_RNDV_LANES=1 UCX_RNDV_THRESH=0 ./src/tools/perf/ucx_perftest_daemon -p 1338

• Run server: UCX_TLS=rc UCX_MAX_RNDV_LANES=1 UCX_RNDV_THRESH=0 ./src/tools/perf/ucx_perftest

• Run client: UCX_TLS=rc UCX_MAX_RNDV_LANES=1 UCX_RNDV_THRESH=0 ./src/tools/perf/ucx_perftest -t ucp_am_bw -n 1000 -s 2097152 -Z ip_dpu1:1338,ip_dpu2:1339 ip_server where ip_dpu1, ip_dpu2 and ip_server are the real IP addresses of the corresponding cluster