Facebook, three read tools on the user token

[fajarhide]

Adds the personal half of Meta: get_me, list_my_posts, list_liked_pages, as a manifest.

All three read through /me, so the single user token a grant already stores is the only credential involved. That is the whole reason this half is buildable now. Pages and Instagram are a separate use case in Meta's console (confirmed 2026-08-17), and each Page there carries its own token that the executor cannot swap per call. They stay out until that capability is designed.

Two decisions worth naming.

No version in any path. Meta applies the app's own default when the path carries none. Pinning a version from memory risks naming one already retired, which would fail every call for a reason the error does not explain. A test asserts no /vNN.N/ appears in the manifest, so pinning it later is a deliberate act.

No longLivedRefresh. The code exchange is traded once for a sixty day token through fb_exchange_token, the same shape Threads uses. Facebook has no th_refresh_token equivalent for a user token, so a grant dies at sixty days and reconnecting is the only path. Its absence is the accurate model, not an omission.

Paging reuses the Meta cursor shape Threads already proved, including the trap where the cursor survives on the last page and only paging.next says whether more exists.

$ npx vitest run
Test Files  36 passed (36)
     Tests  366 passed (366)

Maturity is experimental, and that is the honest label: no request in this file has reached Meta. The fixtures are written from the documentation, which catches a wrong shape and cannot catch a wrong document. Promotion to beta needs a real connection, which needs a tunnel and FACEBOOK_CLIENT_ID, and stays a human act.

Closes #20