So, here comes yet-another token redesign. Thing is, I've found a great Djot parser at <https://github.com/hellux/jotdown?tab=readme-ov-file>, and run into a problem: they produce tokens containing string references already, which does not work well with my previous design based on separate handling of source and tokens.

Futhermore - there's no REAL reason to do that, I just wanted to be fancy or th :idk:

So, here comes yet-another token redesign. This time, tokens DO contain some-sort-of reference to underlying code, but actual TYPE of this reference is being abstracted on. This way, I would be able to use [bytes](https://docs.rs/bytes/latest/bytes/)-based references, `&str`-based, or whatever crazy `Span`-like stuff I'll come up with later.

Thing is, `jotdown` uses both `&str` and `Cow<&str>`, depending on if the source should be sanitized, so I'll empose a restriction on a source type to be `ToOwned`, and will define my own wrapped around `&str` to resolve to `::beef::Cow<&str>` instead.

One more thing I'd like to account for - partial parsing. Also, I'd like for my tree to be fully-modifiable and easily-redactable, so there's that :idk:.