Once #106 lands, we have the ability to figure out the PC (and hence span in the source) for each piece of output. When the user clicks in the output window, we should make that set the breakpoint at the corresponding point and thus highlight the appropriate state at that moment in time.

One challenge (we can forego it for now) will be loops -- i.e., we would ideally select not just the right breakpoint (line/column) but the right instance of the breakpoint. We could track that by comparing the amount of text output at any particular time.