add benchmark-findings.md which is a scratchpad for things that stand out to me with the benchmark results

benchmarks/benchmark-findings.md

@@ -0,0 +1,4 @@
+# cross-copy
+- for internode copy, the src-node-engine is only slightly faster than the dst
+- for intersocket copy, the src-node-engine is faster than dst-node-engine
+- using both in tandem can give about 1.4x speedup over only src-node