Only three choices? I can think of another.
Keep it realistic
When I do a usability study that involves observing participants while using a product, I like to get the scenario as realistic as possible. And I go out of my way not to lead the participants.
For example, to I want to assess whether participants can complete a task that requires them to use a command on the Tools menu, the scenario—or instructions—that I ask each participant to read before they begin will not have the word "tool" in it.
For your product, your real users would typically walk through a tutorial. This could definitely be seen as "leading the participant." But I think the key questions in your case are how long is the interval between the tutorial and product usage, and how quickly do they complete the tutorial? Do they skip through it? Are they usually interrupted?
Then consider: could you reproduce or mimic the real situation in your research?
During each research session, perhaps you can start with a tutorial—be sure to rush your participants so they skim, or to deliberately interrupt them, as would occur for real users. If you ask them to do the tutorial immediately, before you discuss the process and before you record anything, then you'll have several activities that you can use to lengthen the duration between the tutorial and the product use:
- "Please read and sign the copyright waiver and non-disclosure agreement."
- "Please fill in this questionnaire" about the professional domain but unrelated to the features you're testing, or provide some other distraction task.
- "Please read your email for 5 minutes while I set up."
And then carry on with the usability research of the product, the study's meat and potatoes that you want to record.
It'll take extra time, so be prepared to compensate accordingly, and schedule accordingly. In my opinion, it's better to test with fewer participants more realistically than to test with more users who have been cued by the tutorial.
To answer your question, I choose option 4, which you did not list. And that is: mimic real usage as much as possible.