Neural Net Go Problems

[TFiFiE]

Your neural net problems fill a nice niche, but I noticed that in some of them, one just has to take back a ko after a threat-response exchange. I first wanted to suggest that these are (optionally) filtered out, but as these just constitute a special case of the numbering of the last few moves being helpful, it might just be better to have an option to start out only marking any possible illegal ko move and reveal the preceding moves as part of the answer (in the same spirit as https://github.com/lightvector/GoNN#some-thoughts-about-history-as-an-input-mar-2018).