Builtins are versioned in Plutus but not in Haskell

`PlutusTx.Builtins.Internal` has

```haskell
consByteString :: BuiltinInteger -> BuiltinByteString -> BuiltinByteString
consByteString n (BuiltinByteString b) = BuiltinByteString $ BS.cons (fromIntegral n) b
```

but that's the old definition of `consByteString`, the new one doesn't use `fromIntegral` and instead fails if `n` doesn't fit into `Word8`. This means that some builtins behave differently on whether we evaluate them in Plutus or Haskell.

From what I was able to gather, people mostly evaluate Plutus as Haskell just for the purposes of testing, but if the behavior is inconsistent, then such tests can be extremely misleading. We should either make builtins versioned in Haskell as well or remove that functionality entirely and require people to always evaluate Plutus code using a Plutus evaluator.