docs: apply loading lazy, decoding async to index.md

[darjeeling]

What

Fixes issue #1085

AI Provenance (omit for human‑authored PRs)

• Tool: ChatGPT + VSCode
• Model: GPT-5.4
• Thread / prompt: how to fix blocking issue of rendering this page in img tags which will render in mkdocs

Pattern

Reviewer Focus

I choose Option 2

it may affect much, but blocking will reduce.

we may need to consider add these kinds to the html head

<link rel="dns-prefetch" href="//img.shields.io">
<link rel="preconnect" href="https://img.shields.io">

Checklist

• Diff contains only changes for this task — no unrelated refactoring or cleanup
• Addresses exactly one issue or feature
• New or changed behavior has test coverage
• This is the simplest viable approach
• AI provenance section removed or accurate