Fix the tests failing on main

[TheGreatAxios]

Diagnosis

packages/cli/test/seed.test.ts — fixed. Genuinely broken, not order-dependent in the test-execution sense. c0b60f90 ("Land in Myra's room as soon as she is live, not when the seed set finishes", CL-6462) deliberately reordered DEFAULT_WORKFLOWS in packages/hub-client/src/seed.ts to deploy assistant before echo (documented in the source comment above the array), but seed.test.ts was never updated and still asserted the old echo-first order. Fixed the two toEqual expectations to match the current, intended order. Full packages/cli suite is green (27/27).

apps/web/test/library-page-selection.test.tsx — could not reproduce as failing. Ran it standalone (3x) and as part of the full apps/web suite via the real test script (bun run build && bun test ./src ./test, the same command CI uses) — 774/774 pass, 0 fail, every time. One real gotcha: this file needs apps/web's bunfig.toml DOM preload, so running bun test from repo root instead of from apps/web/ gives a misleading document is not defined — not the failure the reviewer saw, but worth knowing if anyone else tries to repro. I could not find a shared-mutable-state leak or any other cause. Did not touch this file. If it's still red on a fresh main-HEAD CI run, it's likely CI-environment-specific (parallelism/memory pressure) rather than a code defect — worth a second data point from an actual CI log before spending more time here.

scripts/e2e/chat.test.ts — could not run, static analysis only. Ports 3100/3199 were already LISTEN-bound (another lane's live stack), so per the machine notes I did not attempt to start a second stack. Read through the file for the "run-id mismatch" pattern (an assertion using an id it never captured) but didn't find one on inspection — the one prior fan-out rewrite (#199) already captures echoLocalPart from the actual participant response before polling /workflows/runs/:runId/events with it. I don't have enough confidence from static reading alone to claim this one is fixed or to identify the exact flake; it needs an actual run against a free stack to observe the mismatch and pinpoint the id that's captured too early/late.

What's in this PR

• packages/cli/test/seed.test.ts: assert the current assistant-first DEFAULT_WORKFLOWS order (matches CL-6462 intent, no production code changed).

Verified

• packages/cli: 27/27 pass
• apps/web: 774/774 pass (via bun run test, build+test)
• bun run lint: clean (prettier clean, 0 eslint errors, only pre-existing unrelated warnings)

Not verified / needs a peer

• apps/web/test/library-page-selection.test.tsx red on CI despite passing locally every time I ran it — needs a real CI log to compare against, or someone to try reproducing under CI-like resource constraints.
• scripts/e2e/chat.test.ts — needs an actual run against a free stack; I only did static analysis.

Do not merge — for peer review per the CI-recovery task.