Well for starters recreate the situation and test out different approaches. Thanks to the detailed analysis that can be attempted.
If I understand right, a good next step would would be with eBPF or some type of proxy ignore the forged RST+ACK at the beginning.
Then it would come testing to see if sending a bunch of ACK packets, perhaps with sequence numbers that would when reconstructed could complete the handshake. Trying to send them alongside the SYN+ACK or even before if it can be predicted. Maybe try sending some packets with sequence id 0 as well to see what happens.
If I understand right, a good next step would would be with eBPF or some type of proxy ignore the forged RST+ACK at the beginning.
Then it would come testing to see if sending a bunch of ACK packets, perhaps with sequence numbers that would when reconstructed could complete the handshake. Trying to send them alongside the SYN+ACK or even before if it can be predicted. Maybe try sending some packets with sequence id 0 as well to see what happens.