> It seems the only feasible way is to use an harmonic mixer with a YIG tuned filter.
No. I hope my YIG post didn't have any part in giving you this impression, because it is not correct. The most common spectrum analyzer architecture is high-IF, which does not use a YIG. So much so that if someone refers to "the" spectrum analyzer architecture, they are almost certainly referring to high-IF, rather than YIG -> harmonic mixer. Even YIG based spectrum analyzers only start using the YIG filter above several GHz, so they will use the high-IF architecture for 0GHz to 2-4GHz, since the YIG has a minimum frequency. If you have a YIG you'll want to use it, but if not, high IF is the ticket.
The limiting factor on high-IF typically pops out of LO subharmonics and the inconvenience of dealing with a high frequency IF. Your instincts are correct: these problems have been greatly alleviated with modern components, especially that TI 15GHz synthesizer. I recently did a high-IF design at 14.5GHz and I suspect if you survey components you might land in a similar region. I won't tell you exactly what I bought because it was opportunistic. The listings were one-off. If you cultivate an understanding of the constraints, though, you will be able to spot similar opportunities, even though the specifics will be different.
A tutorial on high-IF is out of scope, but if you search around for "the" spectrum analyzer architecture, you will find it. The telltale signature is that the first conversion goes up in frequency. Which is strange, because the eventual point is to go down in frequency! You do it to avoid spurs/images. Once you understand this motivation, you will be able to put together a bandplan that uses available components, and modern components make higher high-IF designs much easier to accomplish.
Good luck