Improve site performance metrics

[Blargian]

Summary

Various improvements to pull up site vitals:

• adds docusaurus faster (closes Explore docusaurus-faster #3209)
• add defer and async where missing on tags
• add preconnect resource hints for external scripts
• remove client side rerouting and move it to vercel instead (this is better for SEO in any case)
• lazy load code blocks
• extract critical css
• remove some unused styles, fonts etc

Checklist

• Delete items not relevant to your PR
• URL changes should add a redirect to the old URL via https://github.com/ClickHouse/clickhouse-docs/blob/main/docusaurus.config.js
• If adding a new integration page, also add an entry to the integrations list here: https://github.com/ClickHouse/clickhouse-docs/blob/main/docs/en/integrations/index.mdx

[Blargian]

CC @gingerwizard

Implemented critical CSS extraction (critical CSS gets stripped from the bundled css and inserted directly into a style tag inside the head of html pages)... disappointing results. Shown top to bottom - current site (clickhouse.com), vercel preview with various changes (docusaurus faster, lazy loading code blocks, resource hinting etc), vercel preview with added critical css extraction:

So essentially no benefit for added complexity, added build time to inject the CSS into all HTML elements + still need to spend more time on it as rendering looks bad with local development server (yarn start).

Focusing on all other improvements made (excluding CSS extraction):

Loss of points on performance due to total blocking time:

I'm not sure how much control we really have over a lot of these. Reducing page sizes will help, but in the case of settings, functions etc it will take a big effort to restructure. Pages like this get flagged as good pages, and the only noticeable difference I can see is that the page is shorter.

The size of the main.js bundle is directly tied to how many routes we have. This is a known issue with docusaurus, and something they will improve over time: facebook/docusaurus#7256

Performance improvement from around 40% to around 70% is already substantial + issues with FCP, LCP, CLS are fixed. I'd recommend we revert the CSS extraction changes and stop here for now, validate the fix on search console and see in 28 days if we need to further improve performance. Thoughts?

[gingerwizard]

@Blargian my only concern is rendering issues. Painfully we probably need to sample a decent number of pages. Otherwise, if the improvements are significant merge and do further work under a new PR.