feat(docs): scaffold fumadocs next site (phase 1)

[pratyush618]

First phase of the docs migration off Zensical. Sets up docs-next/ (Fumadocs 16 + Next.js 16 + MDX, static-export preset) alongside the existing Zensical site. The Zensical site at docs.byteveda.org/taskito keeps shipping unchanged; nothing is removed or repointed in this PR.

Replaces #110 (Docusaurus draft) — closed after revisiting the GH Pages constraint. Fumadocs' static export + built-in LLM routes give us all of phase 6 for free.

Why Fumadocs

The original plan recommended Fumadocs for four reasons; you initially asked for Docusaurus on GH Pages, then I confirmed Fumadocs static-exports cleanly to GH Pages. Picking Fumadocs back up because the LLM routes + Tailwind-first DX + +next+fuma-docs-mdx+static template wipe out three of the seven planned migration phases on day one.

What's in here

docs-next/ — Fumadocs scaffold with +next+fuma-docs-mdx+static template, Biome linter, Orama search, Next-OG image generator. Versions:

• next 16.2.4
• fumadocs-core / fumadocs-ui 16.8.5
• fumadocs-mdx 14.3.2
• react 19.2.5
• @orama/orama 3.1.18

Static export config — next.config.mjs ships:

output: 'export',
basePath: '/taskito',
images: { unoptimized: true },

GH Pages-ready out of the box. Output goes to out/.

LLM-friendly routes already wired by the Fumadocs template — these were "phase 6" in the original plan, free here:

• /llms.txt — index of every page with title + description
• /llms-full.txt — full corpus as one Markdown document
• /llms.mdx/docs/[[...slug]]/content.md — raw MDX per page (replaces the hand-rolled copy-markdown.js)
• /og/docs/[...slug]/image.png — auto-generated per-page OG image via next-og
• /api/search — local Orama search endpoint (no Algolia signup)

Taskito branding — src/lib/shared.ts (appName + gitConfig pointing at ByteVeda/taskito), navbar links array (Getting Started / Guides / Architecture / API / Changelog), root metadata with metadataBase: https://docs.byteveda.org/taskito.

Custom landing page — src/app/(home)/page.tsx rewritten with Taskito hero (title + tagline + Quickstart/Install CTAs) and a 3-card feature row (Brokerless / Rust-powered / Python-native). Real visual polish is phase 5.

IA skeleton (33 stub MDX pages + 14 meta.json files):

• getting-started/ × 3 (installation, quickstart, concepts)
• guides/ × 9 sub-categories (core, reliability, advanced-execution, operations, observability, resources, workflows, integrations, extensibility) — folds in Resources, Workflows, Integrations from the Zensical top nav per the approved IA
• architecture/overview.mdx
• api-reference/overview.mdx
• more/ × 4 (examples, comparison, faq, changelog) — collapses Zensical's Changelog + Changelog Archive split

Each stub has a Fumadocs <Callout type="info"> saying "Phase 1 stub — content port pending" so it's obvious nothing is final.

.github/workflows/docs-next.yml — build + typecheck + lint on PR (path-filtered to docs-next/**). No deploy yet — cutover lives in phase 7.

What's NOT in here (next phases)

• Content port (phases 2–4) — actual prose moved from docs/ into the stubs
• API reference component (<APIMethod> — phase 3)
• Landing page polish + shadcn-style components (phase 5)
• LLM-friendly routes ← already in template
• Hosting / DNS cutover and Zensical decommission (phase 7)

Phase 6 gone, schedule shrinks

Original critical path was phase 1 + 6 + 7 (~3 days). With phase 6 absorbed into the template, it's now phase 1 + 7 (~1.5 days). Phases 2–5 still parallelizable post-cutover.

Test plan

• pnpm install --frozen-lockfile — clean
• pnpm types:check — ✓ Types generated successfully
• pnpm lint — exit 0 (2 warnings on Fumadocs scaffold's !important CSS, intentional, no errors)
• pnpm build — 66 static pages, no warnings, no errors
• out/llms.txt and out/llms-full.txt populated with the IA tree
• Static export hits /taskito basePath; out/index.html and out/docs/... generated
• CI: new Build Docs (Next) job passes on this PR
• Manual smoke once merged: spin up pnpm --dir docs-next dev locally and confirm all 5 nav buckets resolve, dark mode works, /llms.txt and /llms-full.txt return content