I'd be more worried about skew. CAT5e and 6a seem to have the same delay skew spec, and I can easily imagine Ethernet PHYs that are simply unable to operate beyond 100Mbps with out-of-spec skew.

CAT7 seems dubious (not recognized at all by everyone in the industry) but has a tighter delay skew spec, so this would help.

Another potential issue would be if the PHY has an equalizer that can handle reflections due to not-quite-perfect termination at 100m or closer but has trouble beyond 100m.

Something like a 1000BASE-T-LR or something that just means "1 Gbps over CAT [whatever is needed] if properly installed at lengths up to 200m" would be quite nice. Most likely a lot of equipment would meet that spec out-of-the-box.

Not sure whether this still holds for 1Gb and faster ethernet, but there used to be a maximum length due to collision detection.

An ethernet frame has a minimal length, and since the cable (used to be) a shared medium, frames could collide. Such a collision could be detected by senders whilst sending the frame, allowing for a re-send. However, when the cable is long enough, you could get collisions happen after you are finished sending.

I'm guessing that the newer standards that only allow point to point connections invalidate those requirements though.

That's not actually true. When the 100 meter limit was established, a minimum-length frame would be 10 kilometers long.