Speed up tokenization

[rien]

The tokenization performed by tree-sitter can be slow for large datasets: up to 50% or even more is spent on this part.

Multiple improvements are possible:

• Try out similar parsers, possibly those that can be run in browser contexts as well, possibilities:
  • https://github.com/microsoft/vscode-textmate - code highlighter used by vscode, requires WASM (but no native deps)
  • https://github.com/highlightjs/highlight.js - a full JavaScript highlighter
  • https://github.com/PrismJS/prism or https://github.com/tannerlinsley/reprism - pure JS as well with support for a lot of languages, but no longer maintained
  • https://github.com/syntax-tree/unist - gestandardiseerd protocol voor syntax trees?
• Parallelize tokenization using workers
• Benchmark/profile tree-sitter, we might just be using it wrong?