I assume you're equating picking a random SSH port with scanning for an ASLR slide or guessing a stack cookie, but they are different situations: processes that die are generally treated quite seriously, and they leave "loud" core dumps and stack traces as to what is going on–usually this gets noticed and quickly blocked. With SSH you can generally port scan people quickly and efficiently in a fairly small space (and to be fair, 16-bit bruteforces are sometimes done for applications as well, when possible)–and the "solution" here where you ban people that seem to hit you too often is literally what you are supposed to be running in the first place.

And in general, the sentiment was "if you are using those things, you are likely to have invested time into other tools as well such as static analysis or sanitizer use" which are not "security through obscurity" in any sense, whereas the "obscurity" that gets security people riled up is the kind where people say things like "nobody can ever hack us because we changed variable names or used something nostandard" because it is usually followed with "…and because we had security we didn't hash any passwords".

How so? Stack cookies and ASLR are a form of encryption, where an attacker had to guess a random number to succeed in an attack.

Obscurity really just boils down to a secret that doesn't have mathematical guarantees. It's doing something that you think the attacker won't guess, just like an encryption key, but without the mathematically certified threat model, so you just hole that the attacker is using a favorable probability distribuy for their guesses

The attacker who had already compromised the integrity of the system in question has to guess or probe for a random number with relatively low entropy in order to do something useful and straightforward with that already compromised system.