TCP injection attacks are not the same as man-in-the-middle attacks... but rather are categorized as man-on-the-side. The difference is important because MoS is *much* cheaper for these various (not just NSA) entities to execute. MoS means you do not have to pwn a route endpoint at the site of your TCP injections... you can inject from almost anywhere as long as you can win the race.
I will discuss this point in my write up... and I will write a section specifically for Tor exit relay operators who are interested in using HoneyBadger.
What about the approach of detecting/preventing those attacks at the user endpoint. Like enforcing HTTPS-connection (HTTPS-Everywhere) and prohibiting/announcing redirects.
Tor users will not be able to detect these attacks on their infrastructure; hence my message to Tor exit relay operators.
It is possible to add a "prevention" mechanism to HoneyBadger; an event based firewall ruleset generator made to block TCP injection attacks as they are happening... yes. This is possible. I could write that if there was interest from enough people.
Yes... users of the Internet should give up using plain-text protocols to stay safer. HTTPS-Everywhere and the various other related efforts by the EFF are all a great help towards keeping people safer.