Yes, I did see the pic. I have done exactly what you are describing with a 7 series FPGA (I can’t afford the one you have
)
Using 1 TX lane and 1 RX lane of GTY you can do 10GigE with an SFP+ module. You don’t need an external PHY (this is half done in the module, half in logic). You literally just wire them to the module. You also need to tie the other pins high or low as I mentioned and provide power to the module.
You then use the Xilinx 10G Ethernet PCS/PMA and you now have a working XGMII interface in your logic. If you have a license for it, the 10G ethernet subsystem IP also includes a MAC and some sort of AXI memory map to packet conversion.
Only catch is that you must have a low jitter clock, and at least for the 7 series devices, it must be 156.25MHz. I’m not familiar with the clocking structures in the Ultrascale devices.
Also if you have 100GigE QSFP cages and they are unused, why not just use a QSFP+ to SFP+ adapter?