docs: standardized page/page_size pagination for service agents

[Alexgodoroja]

What

Documents the new fleet-wide pagination on service agents (styled + plain docs).

• New "Pagination" section on /docs/service-agents: the standardized page / page_size filters (translated per-upstream to page/offset/cursor), bounded page merging (up to 5 pages / 500 records), and the pagination block that carries has_more, total, and the exact ready-to-run next-page command (pilotctl send-message <agent> --data '/data {…}').
• Removed the stale reply-truncation note — it told users replies are "capped at ~8 KB" with … (truncated, N bytes total) spliced into the JSON and to "pass a limit filter." That behavior is gone: agents now return full, valid JSON and merge pages.
• Added the section to the on-page TOC.

Verification

• npm run build passes (364 pages, no errors).
• Changes live inside the bodyContent template literal; braces/tags balanced.

Docs review / sensitive-data check

No secrets or API keys anywhere in src/. One pre-existing item to flag (not touched here, functional bootstrap infra): the rendezvous/beacon endpoint appears as a raw public IP 34.71.57.205:9000/9001 in plain/docs/troubleshooting.astro and plain/docs/webhooks.astro — consider replacing with a hostname in a follow-up.

[github-actions]

🔄 Plain twins auto-regenerated

The machine-UI mirrors under src/pages/plain/ had drifted from their marketing/docs sources. I regenerated the stale page(s) and pushed a commit to this branch:

src/pages/plain/docs/service-agents.astro

Please review the bot commit. Prose is an LLM summary of the source; commands, flags, and numeric specs are copied verbatim. If the build / check:plain check is still red, re-run it (or push any commit) — the regenerated twins now match their sources.

[TeoSlayer]

Holding this one pending verification of the feature it documents. Live test today: gbif-species still returns the ... (truncated, N bytes total) splice this PR removes, and list-agents accepts page/page_size (echoes them in filters_applied) but does not honor page_size (returned count=50 for page_size=3) nor emit a pagination block. So the claim 'Every specialist accepts page/page_size … the agent fetches and merges up to 5 pages … the reply's pagination block tells you how to continue' isn't matched by the deployed specialists.

If the merge/pagination wrapper is rolled out fleet-wide, happy to merge — but as written it replaces a true caveat (8 KB truncation) with a capability that isn't live everywhere, which is the opposite of the accuracy intent. Sibling PRs #139/#140 (privacy scoping, pricing) are verified-correct and merged.