You, probably like myself, took way too long to discover them because they have a fairly niche usage and are probably Postgres' least appropriately documented feature. It makes it hard to imagine the ways that it can be used. We should probably do something about that.
I read the second description in the manual and amazed it exists. I can totally see use cases where you want "almost unique" constraints (same name, date of birth too close??)
In addition, I find the new hCaptcha used by Cloudflare much worse than reCAPTCHA. It's slower, with ambiguous inputs demanding more mental effort, and sometimes it just refuses my correct answers.
Google probably has a good enough handle on you to serve the easy captchas. But keep in mind that for users they can't fingerprint (e.g. with Tor Browser), reCAPTCHA is usually insanely hard or even blocked ("Your computer or network may be sending automated queries"). hCaptcha has been heavenly compared to that.
I've seen some sites suggest they CAPTCHAs will go away if I disable my ad blocker[†], so if not everyone is getting it this could be a factor. Before I bother even clicking to see if I get the CAPTCHA, is the content worth the effort?
[†] I can see how this replaces ad revenue at all, so I assume it is an attempt to make running an ad blocker seem like an inconvenience rather than a benefit.
I've seen alternate used in other places to refer to candidate keys that are not currently the primary key, though I'm fairly sure it isn't a defined standard term.