Using `_BitInt(128)` with `double` conversion generates call to unimplemented builtin for 32-bit targets

Converting a 128-bit `_BitInt` to a `double` on 32-bit targets emits a call to the builtin `__floattidf`, which isn't implemented there:
```
undefined reference to `__floattidf'
```

Consider implementing it there, too. Or, generate generic code as is done for even larger integers.

Example for `_BitInt(128)` vs. `_BitInt(129)`: https://godbolt.org/z/chf3195W1