Add CLI Ghost content type selection and guarded migration

[aileen]

Parent

• PRD #213 — Deliver the maintained HTML-to-Algolia extraction pipeline
• User stories covered: 30–36, 41–43, 48–52

What to build

Deliver the complete CLI path for selecting posts, pages, or both and safely migrate an existing mixed-use Algolia index from legacy slug-managed records to stable Ghost content identity. Preserve posts-only defaults and partial-run safety, reconcile only package-managed records on complete runs, and publish the operator-facing migration release before incremental handlers change.

Acceptance criteria

• The CLI accepts a top-level nonempty, duplicate-free contentTypes array containing only post and/or page; omission remains posts-only and there is no CLI override.
• Posts are fetched first and pages second through independent pagination streams with the existing delay and complete-run page size.
• --limit remains restricted to integers 1–100, --page still requires --limit, and either flag makes every selected-type run a partial upsert that never deletes unseen records.
• Existing slug skip/ignore behavior remains slug-wide across selected Ghost content types.
• A complete run validates and preflights every new record, reconciles stale package-managed records and newly disabled content types, saves the selected set, and leaves unrelated records untouched.
• Same-slug posts and pages coexist through protected content ID/type fields and distinct grouping by content ID.
• Before in-place migration, existing records are classified as valid new managed, exact legacy managed, unrelated, or ambiguous; ambiguity and partial runs against an unmigrated index fail before writes.
• The operator contract requires pausing Netlify writes, copying the index, preflighting, running the complete migration, verifying, and only then deploying identity-aware handlers; required identity settings are applied after record reconciliation.
• The post-migration gate proves valid managed identity/type, fetched census parity, no stale legacy/continuation records, disabled-type removal, same-slug coexistence, unchanged unrelated records, and required facets/distinct settings.
• When in-place classification is unsafe, the documented alternative builds and validates a fresh index and switches configuration.
• Offline process-level acceptance covers all selection, pagination, partial/complete, classification, reconciliation, settings ordering, failure, backup/rollback, and unrelated-record cases.
• A scoped CLI minor release is published and verified with maintenance window, backup, default, migration, settings, compatible Netlify floor, verification, and rollback instructions.

Blocked by

• #224 — Release dormant Ghost content identity primitives