Rewrite using tree-sitter

[krlmlr]

Seems feasible, powerful, and fun. Could become a reformatter for roxygen2 code too, with optional sentence wrapping using https://github.com/segment-any-text/wtpsplit or another library.

@TymekDev: Are you open to sharing your approach and code for tree-sitter-roxygen2?

[TymekDev]

Sure! I don't really have any code to share as I barely scratched the surface. However, I can share some insights.

1. I think the grammar should work similarly to JSDoc. This way, the leading part until #' gets omitted and we are left with repeated @tag <arguments> structure.
2. The grammar needs to take into account that tag arguments can span multiple lines.
3. Implicit title and description (and details?) seemed to be tricky to handle. I would start with handling explicit tags, and eventually go back to handling those as a special case.
4. There are a few different tag categories that I would distinguish: no arguments (e.g., @noRd, @export), one argument (e.g., @import, @inheritParams), and multiple arguments (e.g., @param, @importFrom).
   1. Perhaps it would be pragmatic to start with just no arguments and some arguments.
   2. Eventually some of the tags could have more detailed grammar. For example: @importFrom could distinguish the package name and functions; @param could have multiple variants for "x" and "x,y", and the description could have markdown injected.
5. I remember I was exploring how roxygen2 parses things and I have a vague memory of shenanigans around "invalid" things like #' @noRd something. I am not sure if the tree-sitter grammar should concern itself with it in any way, though. I guess the bottom line is: how many edge cases which roxygen2 lets slip the grammar should reflect?

Overall, I would imagine the grammar to be a number of repeated tag nodes where each node either has no arguments or some arguments—this way, I think, we would end up with node starting with @someTag and ending just before (or a line before) the next @otherTag.