docs: Teaching goose when to think harder

[DOsinga]

Summary

• add a follow-up to the Agentic State Machines post using the automatic reasoning-effort experiment in feat: select thinking effort with Jev #12237
• explain why reasoning effort may be a better first routing knob than switching models
• show how the experiment exposed reusable state-machine effects without putting Jev inside inference
• add a matching illustrated cover

This is stacked on #12130 so the preview includes and can link to the first post. The post is unlisted while the experiment is still in draft.

Verification

• npm test
• npm run build compiles and exports successfully, then stops on the pre-existing broken /docs/gdk/acp/reference link
• checked the local Docusaurus preview at desktop and mobile widths
• verified all external links return HTTP 200

Related: #12237, #12215

[github-actions]

Documentation preview deployed: https://pr-12267.goose-pr-previews-poc.pages.dev