test: validate manufacturing flow and portable restore commands

[cb8010d6]

Problem and changes

Manufacturing had no current blocking real-HTTP journey. Manual restore still depended on tar inside the MinIO container and PowerShell's SQL text pipeline could alter bytes.

• Add explicit-opt-in, loopback-only synthetic manufacturing journey: engineering upload with separate design/review/release actors, pinned revision, BOM consumption and finished receipt, atomic failed-report rollback, reversal, stable immutable movements and idempotent replay/conflict checks. Disable newly created engineering actors.
• Run alongside stocked-trade acceptance against the existing disposable CI API/services, with separate reports. No extra production infrastructure or bypassed release approvals.
• Restore uses the existing pinned helper, checks archive readability before PostgreSQL changes, stops on SQL errors, preserves binary/UTF-8 bytes and supports Compose selection. No live restore is executed by tests.
• CI explicitly requires pwsh so PowerShell command tests cannot silently skip there.

Validation

Independent Astra source reviews of manufacturing and restore passed. Local root Node tests:24 passed,3 PowerShell skipped (pwsh unavailable locally); shell syntax passed. Exact head 06f9a4dfd9f3b5b27bb01f686c78a60a9bfcabc6 passed all gates in run 36288891032: 835 regular tests (28 scripts, 665 API, 142 Web), 77 PostgreSQL regressions, 13 stocked-trade + 12 manufacturing HTTP steps and 9 actual Chromium cases. PowerShell tests actually executed; zero root-test skips. Earlier failed runs exposed mixed-case actor emails and an incorrect 400 expectation (inventory shortage is 409); both are corrected and failure history remains visible. Actual job logs inspected.

Stacked on #43. This is bounded synthetic manufacturing/command evidence, not full manufacturing costing, production restore, operator sign-off, deployment or resolution of #35.

[cb8010d6]

First real HTTP run 36288350341 correctly failed at engineering-actor login: fixture emails used uppercase MFG-, while user creation preserved case and login normalized lowercase. Stocked-trade 13/13 and all other gates passed; importantly all 3 PowerShell restore command tests executed in CI (27 root tests, zero skips).

Head a793fdaa6cc3405fb331ad43fa71c235335b6ded canonicalizes synthetic actor emails before create/login and adds the targeted regression (4 script tests). Real HTTP rerun is pending; no skipped authentication or release gate. The product's new-user normalization inconsistency is being fixed separately, rather than treating fixture normalization as a product repair.