There definitely are good reasons why Ethernet won out over token ring, but there are scenarios where token ring was better. Before modern bridges, Ethernet could struggle with collisions if a network were too highly utilized - especially if nodes were physically spread out.

As for the diagrams, it can sometimes be confusing when it's not made clear what is being represented. Physical and logical topologies can be mixed star and bus and matched in different ways, and diagrams don't always make clear to which they refer.

Back when token ring was designed normally networks would use coaxial cables for communication. No matter if it ran ethernet, token ring or something else, everybody would share basically a single cable. The cable would have T connectors inserted to connect a computer and the end of the cable needed something to terminate it. It didn't need to be a single line, you could have splits and even a star like design, although there were limitations.

And you are right, any disruption anywhere on the line meant the network would go do. That might be someone removing the termination cap on the end, or simply the line being broken somewhere. However because computers were usually connected using T splitters, it didn't really matter if the computer was connected or not. But the connection not being terminated properly could be an issue. Especially if there was another cable connected to the T before being connected to the computer.

Normally in a room the cable would be laid out like a ring although it usually wouldn't be a closed ring, but instead terminated on one end. This meant each computer would be connected to its direct neighbors, but this wouldn't be an active thing. It wasn't like the computer could only transmit to its neighbors and then they needed to pass it on. It was like a shared line, where everyone could transmit and every computer would receive everything transmitted.

When everything switched over to the regular twisted pair cables we know today, it didn't really change from a communications point of view. Every computer wasn't connected to their neighbors but instead to a hub, but just like before anything anyone transmitted could be received by anyone on the network. It wasn't until much later when things like switches became commonplace and not everyone got all the traffic.

My point is, if you have a shared medium anyway, you can get rid of the MAU by having nodes manage the (virtual) token themselves, basically take limited-time turns based in some order like ascending MAC addresses. You could then wire the cable in any way you want with unlimited junctions, taps, whatever as long as you created a graph where all nodes are connected to each other. The entire point of a token ring is to manage a shared medium (that is, a single pair of wires, either UTP or coax, which can efficiently be wired along the shortest, possibly branching path) because if you have to use a direct connection from every endpoint to an MAU in a star topology, you could just have an Ethernet switch anyway.

Yeah, see the pic in the thread. The "switch" (MAU, Media Access Unit) seems redundant to me though, based on what I read I would expect the network interface cards to create a functional ring on their own over a shared medium. Maybe the old cards for ring-topology networks only worked in that one mode and the MAU made them compatible by pretending they were part of a physical ring, cutting computers out of it if they turned off.