As far as I can understand, neither the “rules for parsing floating-point number values” algorithm nor the If the value of the element is not a valid floating-point number, then set it to the empty string instead statement specify anything about requiring the number to be exposed to JS.

Certainly the requirements (elsewhere) for actually doing anything with HTMLInputElement.value are about exposing the value to JS. But those are separate requirements. The “rules for parsing floating-point number values” algorithm and the “If the value of the element is not a valid floating-point number, then set it to the empty string instead” statement seem to be just limited to the string value. And implementation-wise, all they require is to essentially just run a lexer on the string, looking at the string character-by-character — without converting or otherwise processing the string in any way.