Good point. That's been a source of pain for standard_conforming_strings and bytea_output. Many apps are also - sadly - hopelessly broken with regards to text encoding. (Including, unfortunately, PostgreSQL it's self, but more subtly than the obvious brain-dead behaviour of many apps).
Tom makes a good point. Apps that need to care are better off consistently double quoting.
Frankly if we were going to add any option at all I'd want one to force all unquoted identifiers to ERROR so apps could be easily validated not to depend on case folding behaviour at all. But even that has issues and is probably better done in tooling and static analysis or via a plugin.