parent vs child config dir discrepancies

I hit an interesting issue in the linux.execve-null test while implementing #1667: for -early we check config
file from parent which has HOME set and so we find a config file in ~/.dynamorio; but for late injection we wait until in the child where libdrpreload can't find a config file b/c there is no HOME anymore (this
test clears all the env vars).

Which one is correct?

Should we always create a new env var DYNAMORIO_CONFIGDIR so that all children have consistent config dirs? That could break an app that runs children as different users (though we'll fail to do that properly as we can't hande suid anyway: #1666).