We had that issue at work with email account passwords that could be entered into a browser in UTF-8 but would be sent by email clients on Windows in whatever the default encoding there was, usually not UTF-8.

The server just blindly pushed the bytes it received into the hashing algorithm. It didn’t have any means of identifying the encoding used either way. We “solved” it by showing a warning about the bug when people logged in and entered a password with non-ASCII characters. Many people used a web-based email client anyways so it wasn’t such a huge issue anyways. We didn’t want to force customers to only use ASCII symbols.