Consider modelling spans with RangeInclusive

[414owen]

Currently labels are positioned in the text buffer using a Range<L>.
This means it's impossible to represent a label that runs until the end of the text buffer of length L::MAX, because of wraparound.
I also don't see the point of modelling a zero-width span, so RangeInclusive seems like a much tighter fit for this use-case.
We can make invalid states unrepresentable, here.

[jaens]

A zero-width span just represents an insertion point (similar to a text editor cursor), and those could theoretically be useful when an error is of the kind "something is missing from here, please insert it".

(although displaying that on a terminal is technically equivalent to a width 1 span anyway, since the caret can not be positioned between cells... but for IDE navigation it's still different)