Harden live FRED/BLS E2E against date alignment and network outages

[luokerenx4]

Symptom

On 2026-07-16, pnpm test:e2e passed 31 non-trading cases but failed two live macro-data assertions while src/domain/market-data/** was identical to origin/dev:

1. fred_series multi-symbol returns rows with both columns populated returned rows, but none had both GDP and UNRATE populated on the same date (market-data.e2e.spec.ts:59). The failure reproduced in an isolated rerun.
2. bls_series returns observations + skips missing periods returned HTTP 500 because the machine could not establish TLS to api.bls.gov (ECONNRESET). A direct curl reproduced LibreSSL SSL_connect: SSL_ERROR_SYSCALL.

Reproduction

pnpm test:e2e
pnpm vitest run --config vitest.e2e.config.ts \
  src/domain/market-data/__test__/e2e/market-data.e2e.spec.ts \
  -t "fred_series multi-symbol"

Suspected surface

• src/domain/market-data/__test__/e2e/market-data.e2e.spec.ts
• src/domain/market-data/__test__/e2e/setup.ts
• FRED multi-series date/pivot semantics
• policy for external-network failures in the otherwise non-trading E2E lane

Desired outcome

• Keep deterministic join/pivot behavior covered with fixtures or a controlled provider response.
• Keep an optional live-provider health lane for FRED/BLS.
• Report unreachable upstream hosts as a clear live-lane skip/failure without making unrelated release PRs look regressed.
• Define whether mixed-frequency series should produce sparse rows or an aligned/forward-filled table; make the assertion match that contract.

Why deferred

Discovered during provider/Workspace release-readiness verification. The current change does not touch market-data code, and mixing a macro-data contract change into that PR would obscure both reviews.