Title: Step 3 (correctness evaluator setup) may be ordered before the dataset run it depends on

[annabellscha]

Body:## Summary
In "Step 3 — Set up the correctness evaluator in Langfuse", users configure the
Correctness evaluator and target dataset runs, but the step that actually runs
the dataset comes later in the workshop. So the evaluator setup screen has no
data to map or validate against.

What happens

Following Step 3, the "Set up evaluator" screen shows:

• Results table (filter Dataset = dad-it-support-workshop) says "No results."
• Evaluation Prompt Preview says "No trace data found, please adjust filters or
  switch to not show preview."

With no dataset run yet, there's nothing to map the template variables against
or confirm the evaluator is wired correctly.

Expected

Either:

1. Reorder so the dataset run happens before setting up the evaluator, or
2. Clarify in Step 3 that empty data is expected here — the evaluator runs
   automatically on new/incoming experiments once the dataset is run in the
   later step ("Run on new experiments" is enabled), so the empty preview is
   normal.

Reported by

Workshop attendee (Si Shuo Yang) via Slack #langfuse-workshop-feedback.

[annabellscha]

This one makes sense to me.

In the current flow, Step 3 asks learners to configure the Correctness evaluator before Step 4 produces the first experiment run, so seeing an empty results table / empty preview at setup time is normal.

I don't think we need to reorder the chapter, though. The simpler fix is to call it out explicitly in Step 3 and the instructor notes: if no experiment runs exist yet, the review / preview area will be empty (No results / No trace data found), and that is expected. After npm run dataset:run, new experiment items will be scored asynchronously by the evaluator.

That keeps the teaching sequence intact and removes the "did I wire this wrong?" moment.