token_first_child_in_range, token_last_child_in_range, token_child_for_offset limited at the end of the document

This is another use of the AST as a data source:

Performing tree traversal, I noticed that an empty document will produce a `DOC_START_TOKEN` + `BLOCK_EMPTY`, both with a range of `start = 0` and `length = 0`.

This defeats tree traversal implementations in `token_first_child_in_range`, `token_last_child_in_range`, `token_child_for_offset` that can't descend into empty tokens.

I've been working with this more intensely the past months and to my surprise found that an offset and sub-range based check work well like so:

1. `Does a character range _R_ include offset _O_?` A check for inclusion entails that the range is not empty and the offset falls chiefly inside it: `R.start <= O && R.start + R.length > O`
2. `Does a character range _Ra_ contain character range _Rb_?` A check for subrange overlap will falsely (= my argument) reject the same range iff both are empty. (0..<0) and (0..<0) would not overlap at the moment.

For the subrange check I found that this works fine (again, to my surprise), including equal ranges:

(Ra.start <= Rb.start
&& Ra.start + Ra.length >= Rb.start + Rb.length)

Consequences:

- the current approach can't traverse empty trees, as a `DOC_START_TOKEN` + `BLOCK_EMPTY` with length of 0 cannot be drilled down into;
- the end-of-file position is always past everything: past the document range, and past any of its tokens, so you never know the context of the blinking cursor at the very end