Fix incorrect token offsets for multi-byte Unicode characters

[naaish-adam]

Currently, slicing strings using token offsets fails for multi-byte Unicode characters like é or ₹ because the tokenizer advances the offset by 1 for all characters regardless of their byte length. This PR fixes it by advancing by byte length (char.len_utf8()) instead of character length for offset values.

Example JSON that produces incorrect offsets:

{
  "Café": "Malé",
  "Owner": "Jérôme",
  "Location": "Île de la Cité, Paris",
  "Items": {
    "Espresso": "€3",
    "Latte": "€4"
  }
}

[nzakas]

Thanks for putting this together. I'd like to keep the behavior between the Rust and JS implementations the same, otherwise the unified tests will fail. Do you want to take a look at that?