You would know if they are over a cellular network or checking on mobile.
If someone sends you a youtube link and you hit play, YT knows who you are, both from a network perspective and potentially the logged in user.
If you are using signal in a high risk environment, you should be using it from a system that contains no extra information about you. This is the same posture one should take when using Tor.
Basic opsec.
I don't think these kinds of things are in signals threat model. It is meant? as a message platform for people with nothing to hide?
i don't think you can call opsec basic, since it requires tons of knowledge about technology and techniques adversaries might deploy against you. targets of attacks don't neccesarily have this kind of knowledge.
opsec is _incredibly_ hard for a person not deeply into technology and this type of information. you might argue that you need to stick with certain tools and techniques that are known good, but new vulnerabilities and techniques implemented against you can completely shatter previous knowledge on whats good and bad opsec and still break it despite doing it 'very well'. (like certain darknet markets being closed down due to new vulnerabilities being found in the platforms they use...)
most people who rely on opsec/tradecraft for a living, also rely on teams of people to help them maintain it and validate it constantly... (or eventually fail and get bitten).
you are right though that its unlikely a company or app producer would have a threat model tuned to people who want to hide stuff. those things generally tend to be closed down sooner or later. (encrochat and such services...)