I believe the simplest way to do this is add an Aggregator node that collects the results from both the multispectral process and the panchromatic process. Then connect the Aggregator node to the Build Layer Stack node. However, with the Iterator nodes that you have to process multiple files at once, this has the potential to create one huge layer stack from ALL of your files, which you obviously don't want.

If it were me, I would create a model that doesn't have Iterator nodes, and run the model 7 times (one for each MSI/Pan pair). I created and tested a model for this, and it works fine.

Start with an Input Parameters node. Connect it to Radiometric Calibration (set to TOA reflectance) and RPC Orthorectification. This is the multispectral processing chain. Now add another pair of Radiometric Calibration and RPC Orthorectification nodes. This will be the panchromatic processing chain. Connect that Radiometric Calibration to Input Parameters. When the Connect Parameters dialog appears, select "Add New Input(s)" on the left and connect it to "Input Raster" on the right.

Click the Properties icon on the Input Parameters node. Change one "Input Raster" instance to "Select a multispectral file." Change the other "Input Raster" instance to "Select a panchromatic file."

Now add an Aggregator node at the very end, and connect it to the output of both RPC Orthorectification nodes. Then connect the Aggregator node to Build Layer Stack. When you run this model, it will prompt you for the multispectral and panchromatic datasets. It will process them separately, then assemble them together into one layer-stacked file. Repeat with the other image pairs. This is just one suggested approach.