chore(deps): types 1.49.0, Migrate golden handoff from the package

[ABB65]

What

• @contentrain/types 1.47.0 → 1.49.0.
• The Migrate golden handoff test reads MIGRATION_HANDOFF_GOLDEN from the package; the copied fixture tests/fixtures/migrate-handoff-golden.json is deleted, so producer and consumer test the same document.
• The handoff summary reads plan_hash and order_id when the document carries them (older documents without them stay valid: unknown, never stale), and the sync logs both so a stale handoff left by an earlier delivery can be traced. Comparing against the expected order is separate work after the E2E.

Verification

• pnpm typecheck clean; unit + integration suites: 2523 passed, 1 skipped.
• New test: stamped and unstamped handoffs validate, and the summary carries the ids only when present.
• Studio main CI was green (29904a1) before this push.

[ABB65]

REVISE on ccecf0c (small).

1. The new ids are read from the customer's repo without a shape check. The handoff comes from a file in the customer's repository. validateMigrationHandoff does not look at plan_hash or order_id, and the new code puts them into the summary (client-facing) and into a console.info template as they are. I ran it: { ...golden, plan_hash: { a: 1 }, order_id: 'x\n[fake] line' } validates (null), and the summary carries planHash: {"a":1} and orderId: "x\n[fake] line". The log line would take the newline, so a crafted file can forge log lines.
Fix: accept the value only when it has the shape Migrate stamps. plan_hash is /^[0-9a-f]{16}$/ and order_id is /^ord_[0-9a-f]{24}$/ (Migrate's validateHandoff enforces the same). Anything else is treated like an absent field: unknown, not stale. One small helper used by both summarizeMigrationHandoff and the log line is enough. Please add a test with a non-string and a newline value.

2. The new test asserts conditionally. if (handoff.plan_hash) expect(...) passes silently if the golden ever loses the field, which defeats the point of a contract test. The 1.49 golden carries both, so assert them unconditionally (expect(summary.planHash).toBe(handoff.plan_hash)).

Checked and fine: lockfile diff is only types 1.47.0 to 1.49.0 and pnpm install --frozen-lockfile passes; the golden test and the handoff unit tests pass (8/8); the fixture deletion leaves no other reference. I am not giving a full ONAY yet, so I did not run the integration projects; I will on the revised head.

[ABB65]

ONAY on e3c12c4.

• Both REVISE points are fixed. readPlanHash (/^[0-9a-f]{16}$/) and readOrderId (/^ord_[0-9a-f]{24}$/) gate the summary and the console.info line; anything else (non-string, array, newline, wrong form) reads as unknown. The tests cover 42, an array, a trailing-newline hash and order id, ZZZZ and an empty string. The golden test no longer has conditional asserts: it asserts the 1.49 golden itself matches both shapes, then compares the summary.
• Shapes match what Migrate really writes: its planHash (packages/runner/src/job.ts) is two 8-hex halves, i.e. 16 lowercase hex, and Migrate's validateHandoff enforces the same two regexes. The golden stamp round-trips, so no real hash turns into unknown.
• Lockfile: only @contentrain/types 1.47.0 to 1.49.0; pnpm install --frozen-lockfile passes. No reference to the deleted fixture remains.
• pnpm test:ci projects run separately with long timeouts and --maxWorkers=1: unit 2037 passed / 1 skipped, integration 487 passed, nuxt 274 passed. nuxt typecheck exit 0. eslint . has 0 errors (7 warnings, all in files this PR does not touch). Needed contentrain-query generate in the fresh worktree for the #contentrain client, as with any new worktree.