The offset cache contains an entry for each byte so it can't be accessed via the string length.

Adds tests for all variants except for this:
```
"fo
o" "ba
’"
```

For some reason, this still has the wrong offset.

https://github.com/ruby/prism/commit/a651126458