SGML was/is incredibly complex, formal, top-down, and bureaucratic. It was unapologetically "the book of everything". You needed to do a lot of work to gain some value from this complexity. So a document creator, unaided, wasn't going to bother.
But the protocol on its own doesn't define the future. Capital matters too. If VC money had made different bets in 1994, the communities and individuals writing would have had a different kind of help. WWW was one kind of kludge. Other kludges would have happened too. A mere fraction of the stuff that TBL omitted from WWW could be very empowering to people engaged in communities, needing communication tools, and help with documents. That "help" either never came, or only came via serve-side fiefdoms, like Facebook.
This was not inevitable. What is impossibly difficult one day can become easy, with tools, a few weeks later. We will never know.
"And if search engines were foolish enough to take the markup seriously, it would quickly become worse than useless as the markup became just another SEO toy."
Cause < > effect.
Search engines go where the people go. Gamers need something to game.
If a community of anglers, or Conservatives, or swingers, devise and popularise a language, or a tag set, then the search engine will follow. It will need to understand it. It has no option but to follow.
You ascribe some God-like omniscience and wisdom to search engines that simply isn't there. They are followers not leaders.
All I'm saying is: the web we might have had would look very different to the web we have today. The web we have today is the result of technical incompetence and the prejudices of finance capital, c.1994.
Replay the game, and you get a different result. That's all.