A less heavy handed solution would be a linting tool that linted when public futures are not Send, which authors of libraries could use on their CI and while developing. I don't know if clippy has already considered implementing a lint like this.
Of course there are people who want to multiplex futures over a single threaded executor, for whom this lint would not be useful. But there's a pretty clear ecosystem split between those use cases (which mainly target embedded hardware) and libraries which would want to use this lint.