?\u3042"" produces a broken string on non-UTF-8 input

```ruby
# encoding: Windows-31J
p ?\u3042""
```

```
% ruby --parser=parse.y test.rb
"あ"
% ruby --parser=prism test.rb
"\x{E381}\x82"
```

Note that U+3042 is represented as `\xE3\x81\x82` in UTF-8 and `\x82\xA0` in Windows-31J.