If you run a monolithic, centralized service specifically designed to avoid censorship, and you don't moderate what users do on your service, and some of those users hurt people with your service, then you should expect your service will be shut down as Parler was.
And if you do the same thing, but separate the front end from the back end, and have different entities run them to provide legal separation, while practically and functionally the result is identical to that of the monolithic centralized service and your system is used to hurt people, then you should also expect that whatever components can be deplatformed will be deplatformed.
Your litmus test of 'is used to hurt people' is completely true of web browsers and email too.
Emails are always coming up in court cases etc. as people regularly use them to organise or discuss criminal acts, and it is sometimes used with E2E encryption so nobody can intercept and police the contents. I'm not convinced you've drawn a clear line. When is a protocol client responsible for the content shared or accessed with it, and when is a client not responsible for it.
If you run a monolithic, centralized service specifically designed to avoid censorship, and you don't moderate what users do on your service, and some of those users hurt people with your service, then you should expect your service will be shut down as Parler was.
And if you do the same thing, but separate the front end from the back end, and have different entities run them to provide legal separation, while practically and functionally the result is identical to that of the monolithic centralized service and your system is used to hurt people, then you should also expect that whatever components can be deplatformed will be deplatformed.