If you put that aside, you have a multi band transmitter and receiver in a phone. Assuming you could somehow make it do arbitrary things (which is a big stretch), you'd have a bunch of challenges to overcome.
Firstly, nothing in cellular protocols is really designed to handle uncoordinated random access and users transmitting freely - you'd get collisions and interference when devices transmit over each other. You'd have different devices trying to do different things with each other, so you would need to implement something more akin to WiFi.
Running something more like WiFi on other spectrum would be an option, although you don't (legally) have access to spectrum below 2.4 GHz in most countries that you could use in this way. You'd also struggle to do this anywhere there's a "real" mobile operator using the spectrum, as their base station downlink is significantly more powerful than your handset transmitter.
Finally, one of the reasons mobile networks work over the distances they do (and you cite) is that they are able to use rooftop/hilltop and well-located radio sites with near line of sight over their service area. They also have large antennas focused on a given sector angle, with significant receiver gain - this helps them to receive signals from you when you're further away. This is what helps your phone with its (say) 250 mW maximum transmit power work over kilometres. You wouldn't get any of that in a mesh.
If you tried to build a mesh like this, even putting aside all the software challenges and we assumed it was possible, you'd like find the performance/range not much better than WiFi.
You'd effectively need to run WiFi protocols to handle multi user access on the airwaves, and despite having access to better propagating spectrum, you'll not get the benefits of range due to the small on-device antennas. Meshing WiFi together with some kind of gossip based forwarding protocol at application layer would likely be more effective.