The testsuite should not generate strings with invalid characters

This is funny:

```
> import qualified Data.ByteString.UTF8 as BU                  
BU> BU.toString (BU.fromString "\65534") == "\65533"          
True
BU> BU.toString (BU.fromString "\65535") == "\65533"          
True
```

U+FFFE and U+FFFF are not valid Unicode