chore: adapt cutover-dry-run and operator docs to JSON-based importer

The cutover-dry-run orchestrator was wired to the mysqldump-based importer
with `--sql` + `--private-store` arguments. With the JSON importer in place,
adapt:

  - cutover-dry-run.ts now wraps importLaddrFromJson in dry-run mode and
    compares per-sheet imported counts against the laddr server's reported
    `total` for each list endpoint. Tolerable-diff thresholds carve out
    known data-quality drops (tags with no resolvable namespace, http-only
    buzz URLs).
  - cutover-dry-run.test.ts uses an in-memory fetch mock instead of the SQL
    fixture (which was deleted with the mysqldump-parser removal).
  - docs/operations/cutover.md drops `--sql` from every command and rewords
    the T-3, T-1, and T-0 steps to describe pulling from the live laddr
    site and committing snapshots on the `legacy-import` branch.
  - docs/operations/cutover-rollback.md updates the read-only-source line.
  - specs/architecture.md rewrites the "Data migration" section to
    describe the snapshot/merge model rather than "one big commit."

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: themightychris