CI: run database-backed package suites in walking-skeleton

[TheGreatAxios]

Fixes CL-7120 — https://linear.app/abklabs/issue/CL-7120

Problem

Every DB-gated suite under packages/* and apps/hub/src gates on
e2eDatabaseUrl() (scripts/e2e/harness.ts:27-38) and describe.skips
without a database; their headers claim E2E_REQUIRED=1 turns that
into a loud failure. In .github/workflows/ci.yml, build-test (the
only job running bun run test) has no Postgres and never sets
E2E_REQUIRED; walking-skeleton (the only job with Postgres) only
ran scripts/e2e, test/isolation, and apps/hub/test. As a result
~39 test files skip green on every CI run.

Change

• Add a step to walking-skeleton, after the hub DB suites step, that
  runs every DB-gated suite (grep -rl DATABASE_URL --include='*.test.ts' packages apps, excluding apps/hub/test which already runs) with
  E2E_REQUIRED: "1" against the job's existing Postgres service.
  Discovering suites by the same DATABASE_URL signal they gate on
  means a newly added suite is picked up automatically, with no
  hardcoded file list to maintain.
• build-test is unchanged and stays the no-DB unit gate.
• Note: apps/hub/src/artifacts-mount.test.ts matches the DATABASE_URL
  grep and will run in the new step, but it isn't actually DB-gated — it
  mounts against a stubEngine, not a live Postgres connection. It's
  harmless to include (no real DB dependency to fail on) and excluding
  it would need a special case, so it's left in.
• This is a CI-only change; there is no new unit test to
  write beyond exercising the existing suites against a real database.

Tests

Ran the exact suites the new step will run, locally against Postgres:

E2E_REQUIRED=1 bun test $(grep -rl DATABASE_URL --include='*.test.ts' packages apps | grep -v apps/hub/test)
# 126 pass, 0 fail, across 39 files, ~9-70s wall time

The existing walking-skeleton job runs in ~4.5 minutes; the added
step adds well under a minute, staying inside the job's 20-minute
timeout.

[TheGreatAxios]

Dropped the needs-you.test.ts gating commit: #424 deletes that file, so the edit was moot and only produced a merge conflict with the re-pin stack. The CI change stands alone.