Under FAF we now wrongly accept `final $($expr)?` in stmt position parsing it as `$($expr)?`

[fmease]

Found thanks to @matthiaskrgr continuously fuzzing https://github.com/fmease/rasur versus rustc (downstream report: fmease/rasur#26).

---

With the introduction of FAF in PR #151783, we now wrongfully accept the following program (if the feature is enabled):

#![feature(final_associated_functions)]

fn main() {
    final;
    final 1 + 1;
    final { println!("text"); };
}

It essentially gets parsed as:

#![feature(final_associated_functions)]

fn main() {
    ;
    1 + 1;
    { println!("text"); };
}

[fmease]

cc @mu001999

[mu001999]

@rustbot claim