This test fails because the hid_listener_keycode_released function 
releases implicit modifiers always, even if they were not set by the
key that's going up. Also see the comment in that function:

    If LC(A) is pressed, then LS(B), then LC(A) is released, the shift for B will be released
    prematurely. This causes if LS(B) to repeat like Bbbbbbbb when pressed for a long time.
    Solving this would require keeping track of which key's implicit modifiers are currently
    active and only releasing modifiers at that time.