And if someone mounts a camera in the smoke detector in my apartment they could see me type in the double super secret password whose plaintext has never been stored by a computer.

So what's your point, exactly? Tools which generate good passwords and store them locally on your computer with decent-enough security are light-years ahead of what most people do, and their use should be encouraged.

But in my experience, too many people confuse "this is a possible threat" with "this is a threat it's reasonable for me worry/take action about", and that's what I was getting at.

1password is easier and a much juicier target, so compromised binaries are definitely a possible threat. How would you know it was compromised, unless you had a packet sniffer going on your network?

It's possible that a password manager could be completely open source, distributed with checksums and public key signatures, that my C compiler could be completely open source and distributed with checksums and public key signatures, and I could still end up with a backdoor in the final compiled result, detectable only by deciding to carefully examine the binary I just built.

This is a "possible threat". Should I use that possibility as a justification for never using a password manager? Or for never using a C compiler?

And it's not just a "possible threat" but a likely one, and it's happened multiple times in the past. Sites get hacked all the time, and trojans are standard practice these days (unlike compromised-binary-generating C compilers):

https://threatpost.com/en_us/blogs/savannah-gnu-site-comprom...

http://www.esecurityplanet.com/trends/article.php/2248811/GN...

https://threatpost.com/en_us/blogs/apache-site-hacked-throug...

What your suggesting is far less complex than the Stuxnet virus. You need to get binaries from reliable source and verify checksums whenever possible. You can't protect yourself a 100%, but that doesn't mean you should leave or your doors unlocked.