Switch to Vite for local dev and deployment

[SLaks]

See #121

Note: The individual commits in this PR do not work without the entire chain; I divided it into multiple commits just to make the changes easier to review

• Import CSS from JS so that Vite processes it correctly.
  • Import the TS entrypoint and let Vite handle bundling.
  • Replace all use of fetch() with dynamic import().
  • This means we don't need to worry about building, copying, or deploying the data files; Vite bundles them into separate import()ed JS files in the optimized build.
  • This build takes ~3 seconds on my machine to process all of the JSON.
  • This does have one downside: Vite generates a map from the original import()ed JSON filenames to the generated JS asset filenames, which is why the main bundle is now 292KB (66KB after gzip)
  • OTOH, this makes offline mode much simpler; we can rely on Vite to handle all URLs to cache.
  • Consolidate the pacakge.json scripts to use Vite

[SLaks]

This also removes 2 unused PNG files

[SLaks]

AFAICT, the GitHub action should work for future PRs after this one is merged.

pull_request_target runs on pull requests, but uses the action config from the target branch in your repo (otherwise, it will fail because my action config cannot have access to your repo)

[SLaks]

BTW, replacing fetch() with import() also has other benefits:

• import()ing JSON files is strongly typed, so we'll get compiler errors if the JSON doesn't match the type you consume it as.
• import() is automatically cached, so we can remove the explicit schedule.json cache.

I will probably send a followup PR to take advantage of these benefits.

If you don't want the code size cost, I should be able to use import.meta.url to make Vite correctly resolve (and I think deploy) the JSON for fetch(). However, that might make offline a bit more annoying to build.