Offload gzip compression and bucket loading/decoding/serializing to web workers

[philippotto]

URL of deployed dev instance (used for testing):

• https://comlink.webknossos.xyz

Steps to test:

• test the components which are related to the webworkers I created:
  • general bucket loading
  • 4-bit mode
  • saving (skeleton as well as volume tracings)

Issues:

• fixes Investigate whether bucket loading should/could be done in a webworker #2967

---

• Updated changelog
• Ready for review

[daniel-wer]

Awesome, I couldn't find any errors and the screenshot tests ran successfully as well! 🎉 It also "feels faster"™

What I would really like to do, are proper benchmarks, but I understand that this is hard. We do know, that the gzipping and base64 operations are computationally expensive, though, so moving them off the main thread should definitely have a positive impact! I think this PR is good to merge (unless you can think of a nice benchmark that we could execute before) :)

[daniel-wer]

Wow, awesome that you took the time to write this explanation, extremely helpful and easy to understand! 🥇

[philippotto]

Awesome, I couldn't find any errors and the screenshot tests ran successfully as well! tada It also "feels faster"™

Nice, I'm glad to hear that!

What I would really like to do, are proper benchmarks, but I understand that this is hard. We do know, that the gzipping and base64 operations are computationally expensive, though, so moving them off the main thread should definitely have a positive impact! I think this PR is good to merge (unless you can think of a nice benchmark that we could execute before) :)

Yeah, proper benchmarks would be really cool, but I'm afraid it's very hard to achieve. In the case of bucket loading, this PR is an actual improvement regarding time, but usually web workers would only account for better responsiveness. Time-measuring benchmarks couldn't say much about that.

Nevertheless, we could have some sort of integration-benchmark which loads the tracing view and measures the time until the view is complete. However, I'm not sure whether puppeteer and co. are representative for performance 🤔

---

I pushed another commit which uses flow to enforce that web workers are always properly instantiated after import. That should give us some extra safety :)

[philippotto]

Maybe we should add a linting rule for that, too ^^