it is another hoop to jump through
Perhaps the cleanest option from a user experience would be that manually provisioning a slave via the UI or cli would always provision new instance, regardless of stop vs terminate idle configuration.
For users with stop on idle, the automatic codepath based on queue demand could restart stopped nodes. The stopping and starting is done automagically based on demand. You idle stop and you restart based on demand. The cli has disconnect-node and reconnect-node commands, so that seems covered too.
If Jenkins is configured with stop on idle and you want to hide an ec2 instance it from Jenkins, manually updating ec2 tags doesn't seem too onerous. You simply revert back to the original tags to put it back under management. Wanting to temporarily assert control over the node and have Jenkins leave it alone is akin to temporarily detaching a running EC2 instance from an ELB or an ASG. It is a managed machine, and you need to take explicit action to remove it from management.
With the above, the in-context help is pretty easy to explain.