Implementation is the hard part, as ever.
Take a look at how your desktop internet browser does it:
you can navigate in a single tab forever by following links. At some point, you want to break your flow to split it between to things. You can create a new tab and start doing something in it (even browsing the same websites as the first tab), and you can also follow a link from the first tab in a new tab (middle click).
If you see it that way, it is not difficult nor a burden to the user, as it is decided on the fly with a single action :
* Do whatever you want, you'll stay in the same task flow.
* You want to break from this one ? open in a new flow (long press on the actions?).
* You want to start something else ? create a new flow (a bit like the app launcher while you are in an app).
* Finished with this flow ? close it (swipe down from example).
* Need to switch to another already running flow ? go to the multitask view
(like current left/right swipe)

It has to be done on the fly, not with tagging or virtual desktops where you have to move windows.

But to see it like that we first have to forget the concept of app (where an app has several view, like phone has call history/enter phone number/calling view), and extract all the views to be used by anything/anywhere. At least with some compatibility from one view to the next, like when a picture is taken is can be sent or edited, but not navigated to (geo-tagging of pictures could allows that thinking of it...).