Design Goals

This PR implements the Ambient CSS import column of the table below. The ?url import + Explicit API column is part of a separate ongoing progress.

Included Fixes

• Broken Vite CSS HMR and auto load is fixed (regression from devinxi)
• ?url imported asset's are not server rendered anymore
• Fixes [Bug?]: CSS is not applied on build when Vite build.cssCodeSplit: false option is used #1324
• Fixes [Bug?]: Required CSS is removed when components are re-exported #1843
• Fixes [Bug?]: The same stylesheet appear above html element when use props.children to render  #1704
• Fixes [Bug?]: Fragmented and imported css is injected into the body tag and causes html to repaint/shift #1654

Limitations

• The lazy transpilation uses a naive approach to narrow down the target files:
  • It can't detect if a file is dynamically imported, so it adds the necessary id$$ export to all jsx/tsx files with export default
  • Lazy loading components from aggregated modules e.g. export { default } from "./xyz.tsx cannot mount the related css, as the aggregated module misses the id$$ from xyz.tsx

New CSS fixture

Tasks

• Create an example for css testing
• ?url imported css should not be server rendered
• <link> should be mounted before suspension-end during spa navigation
• Suspension from spa navigation should only end after <link> has been loaded - Will be part of a separate API
• Assets should be deduplicated on ssr
• Public manifest and asset mount API (Edit: Public manifest API analog to RFC: Client assets metadata API for SSR vitejs/vite#20913 will be implemented in a different branch)
• Integrate useAssets in dev (making dev and prod use the same API)
• Implement unified lazy
• Optimize unified lazy plugin
• Test glob import support for lazy
• Convert assets to tags in the useAssets runtime, instead of the manifest json
• Test CSS Order, is it better? Is it the same?