Instead of this we could also run some simpler logic to iterate over the points, like BKE_gpencil_data_minmax() does. But that's effectively what BKE_gpencil_visible_stroke_iter() does as well, just with additional features that we don't use.

The problem with BKE_gpencil_data_minmax() is that is running for all datablock and for the preview we just need the active (visible) frames.

The best solution would be to create a BKE_gpencil_strokes_visible_minmax() to consider only the active frames. If you agree, I can do it.

My mistake...looking at BKE_gpencil_data_minmax() does only active frame, so this is what we need.

The problem with BKE_gpencil_data_minmax() is that is running for all datablock and for the preview we just need the active (visible) frames.

Are you sure about that? It uses bGPDlayer.actframe for drawing.

The best solution would be to create a BKE_gpencil_strokes_visible_minmax() to consider only the active frames. If you agree, I can do it.

We can't simply use the normal min-max calculation since that is axis aligned. The camera fitting should be view aligned.

What BKE_camera_view_frame_fit_to_scene() does is iterate over visible points and executes a callback for each, we should just extend this iteration code to support the grease pencil case.
What I meant with my earlier comment was, that we don't have to use the (rather complex) BKE_gpencil_visible_stroke_iter() for this. We could do a simpler iteration like BKE_gpencil_data_minmax() does it. The important point is that we can execute a callback for each visible point.