[Migrate Plugin] Answer Recap wait has no persistence point

Related PR: https://github.com/awslabs/startups/pull/183

The recap's mandatory wait has no persistence point — wizard_stage has no recap_pending value and the draft is only written between batches, so a session dying at the recap wait re-asks the final batch on resume, contradicting the orientation's "progress is saved." Suggest writing the draft (with a recap_pending stage) before presenting the recap.

Fix: after the final essential batch (incl. Q27 and the "use defaults for the rest" path) is interpreted, write the draft with all resolved values and `wizard_stage: "recap_pending"` before presenting the recap. On resume at `recap_pending`, re-present the recap from the draft — don't re-ask questions. Same for the Full Flow variant.

AC:
- [ ] Draft written with `wizard_stage: "recap_pending"` before the recap is shown (wizard + full flow, incl. defaults path)
- [ ] Step 0 Case 2 resumes `recap_pending` by re-presenting the recap, not the questions
- [ ] Both plugin copies updated (advisor + migrate), drift check green

Locate the wizard_stage handling and recap flow in both the advisor and migrate plugin copies, including the Full Flow and defaults paths. Check how Step 0 Case 2 resumes, then run the repository's drift check; done means recap_pending is persisted before display and resume re-presents the recap without re-asking questions.

Written by the indexing model from the issue text.