Tree-sitter-based token parsing for polyfont — produces per-token scope classifications for the polyfont LSP.