Something that came up while discussing bug 1533985: we don't have telemetry on the time it takes a child process to reach the
PROCESS_CONNECTED point, which makes it difficult to understand whether a particular timeout for that stage is adequate in practice, or the impact of synchronously waiting for it.
(We indirectly have some info about this on Mac, because
CHILD_PROCESS_LAUNCH_MS includes blocking on the child to send back a Mach message before we can reach
PROCESS_CREATED, but it's reasonable to expect the Windows userbase to have different performance characteristics.)