Migrate docker-compose pages to reference pattern (Rosetta-style)

[dkijania]

Part of #1195 (Track 3). Blocked on MinaProtocol/mina#18826.

Goal

Rewrite the four docker-compose docs pages to mirror the existing Rosetta page (docs/exchange-operators/rosetta/docker-compose.mdx) — a thin narrative wrapper around canonical compose recipes living in MinaProtocol/mina. This removes the inline YAML and the tag drift that comes with it.

Why this approach (vs. inline + tag check, or build-time embed)

The Rosetta page is the proven model. It has dense, scrapeable content (services table, env-var tables, make targets, verification commands) that describes the contract — not the content — of the deployment. The actual docker-compose.yml lives in mina, which means:

• No drift. Tags, image names, and flags update in one place.
• AI agents prefer it. Modern coding agents (Claude Code, Cursor, etc.) follow GitHub links, clone, and run make mainnet. A self-contained dir + README is the ideal artifact for that workflow — better than inline YAML they have to paste.
• Naive RAG indexers prefer all-in-one, but llms-full.txt already serves that crowd at the bundle level.

Inline + drift-check CI was rejected: every release would re-trigger a paste cycle. Build-time embed was rejected: extra CI complexity for marginal benefit once we have the wrapper pattern.

Pages to rewrite

• docs/node-operators/block-producer-node/docker-compose.mdx
• docs/node-operators/archive-node/docker-compose.mdx
• docs/node-operators/snark-workers/docker-compose.mdx
• docs/node-operators/seed-peers/docker-compose.mdx

Page template (mirror Rosetta exactly)

Each rewritten page should have these sections, in this order, matching docs/exchange-operators/rosetta/docker-compose.mdx:

1. Intro paragraph + link to canonical dir in mina (e.g. mina/src/app/cli/docker-compose/block-producer/).
2. Quick start — git clone, cd, cp example.<network>.env .env, vi .env, docker compose up -d / make <network>.
3. Services overview — markdown table.
4. Configuration — env-var tables grouped by concern (Docker images, Network, Ports, role-specific).
5. Data persistence — bind-mount table.
6. Make targets — table.
7. Verifying the deployment — sample commands.
8. Clean start — make clean + restart.

Strip all inline services: YAML.

Acceptance criteria

• All four pages rewritten to mirror Rosetta's section order and tone.
• No inline docker-compose.yml content remains on these pages.
• Each page links to the canonical dir in mina (URL must resolve — verify in PR review).
• Sidebar labels unchanged.
• validate-docker-images script either dropped for these pages or pointed at the canonical files in mina.
• llms-full.txt regenerated.

Optional CI follow-up (separate PR)

Two cheap drift-detection checks:

• Link liveness — assert every github.com/MinaProtocol/mina/... URL in docker-compose.mdx files resolves to an existing path on the referenced branch.
• Contract drift — parse the canonical docker-compose.yml and .env.example from mina and assert that env vars / service names / make targets named in the doc tables match. Catches MINA_REST_PORT renames and similar.

Both are short Node scripts, run in the same CI step as validate-docker-images.

Blocked on

MinaProtocol/mina#18826 — the canonical compose dirs for block-producer, snark-worker, seed-peer, and archive must land in mina before the docs2 wrappers have anywhere to point.