Multiple testnode can concurrently run the same test, so create test
results only at the end to reduce the risk of race conditions on test
completion leading to tests never reported as completed.

This shows a limit of the current approach; switching to nxdtest
appears desirable.