Generate enum for field and node kinds

[glfmn]

Hello! I am attempting to use tree-sitter as the initial parser for a compiler using rust bindings to the grammar I have generated.

When it comes to translating the AST I realized that I either have to use strings to identify the nodes or manually keep some sort of enum where I track the kinds. I have considered trying to use the tree-sitter api itself or bindgen to attempt this, but I realize there isn't really a good stable number that could be used for such an enum after looking at #2056.

My concern is that if I manually implement the mapping to an enum, the numbers or strings used for comparison will go out of sync as I make grammar changes. I understand that a mapping will still need to be made between the nodekind ids and the tags in my intermediate representation, but if the nodekind(id)s can change out from under me as I add features, it feels like shaky footing to work from.

It feels like this could or should automatically be generated during the build.rs of the tree-sitter-[my-lang] crate. This way there will be compile errors if things change in such a way that the consuming program's mapping isn't correct anymore.