The "general" idea is to construct a FC network with BCB functionality in the backbone and BEB functionality at the edge. This way you create a very stable backbone network, only SPBM/ISIS L2 routing is enabled. Extra advantage is that there is no license needed on BCB switches to e.g. use VRF/L3-VSN because they are not "IP routing aware", they just perform L2 forwarding.
BCB is a switch without UNI ports - it has only NNI ports active - simple as that. If you terminate the user traffic there, then it becomes BEB. On the other hand it does not change much from fabric perspective. You can have the network with BEBs only - in fact this is the beauty and flexibility of fabric connect. The only drawback is that you need to configure services in the core, so it destroys "do not touch the core" concept.
Yes in order to provide any EDGE service on your middle switch you need to have: 1. Some Gix/y port configured (with flex uni enable) - it is your UNI port 2. I-sid created - as you have on the picture (i-sid 200960 elan) 3. Map i-sid to vlan/port (c-vid 960 port Gix/y). Exacly the same as for other BEBs. Alternatively you can connect a device (PC) which does not do the vlan tagging by default:
i-sid 200960 elan (elan:200960)# untagged-traffic port Gix/y In this way your PC will be a member of same L2 domain as the rest devices in vlan 960.
The config you provided on second picture does not make any sense. You just created local i-sid 200960 but no UNI port (vlan/port mapping) is configured. I-SID creation makes sense only when you terminate end user traffic on that particular node.
I don't know how to respond to your "why" question. BCB and BEB describes only characteristics/use-case of fabric node. Fabric node with UNI ports is called BEB as it terminates edge/customer traffic and do the MacInMac encapsulation. BCB is a fabric node which does not terminate any edge traffic - it has only NNI ports and it passess traffic already encaplsulated by BEBs. So you can imagine that BEB is doing more job because it has to encapsulate packets into MacInMac and then pass it to the next node - doest not matter if it's BEB or BCB. Technically the difference is just in configration. You can have multiple BCBs in your network when needed, but as well you can have no BCB in your network - only BEBs. If you think about BCB as some really special fabric node which cannot terminate user traffic - you're wrong.
For spine and leaf architectures in IP fabric, it is true that spine cannot terminate user traffic as fabric node usage is strictly limited by design. But it does not apply to fabric connect (spbm fabric).