fix(universal): stop reporting shows as OPERATING outside park hours

[cubehouse]

Summary

Universal show entities were reading OPERATING straight through the overnight
park closure and never getting rewritten, so the last-known status froze.

Sampled overnight with the park's own venue schedule confirming it shut, most
externally-shown show entries carried a live-looking status (either an
explicit OPEN, or CLOSED/CANCELED with a future showtime still listed —
show_times lists the whole day's slots from midnight, so "has a future
performance" never goes false overnight). Neither signal referenced park
hours at all.

Show status is now clock-gated against the park's own venue schedule the same
way #317 did for SeaWorld: OPERATING (from either an explicit
OPEN/RIDE_NOW or a future showtime) only holds while the park is actually
open. Delay and long-closure signals are untouched — neither claims
OPERATING, so there's nothing to gate.

Hardening after review

This branch went through three independent reviews (crash/correctness risk,
real-world regression risk, test coverage) before being opened as a PR. They
found real gaps, now fixed:

• isParkOperatingNow could throw on a malformed-but-successful schedule
  response (a non-array body), taking down buildLiveData for the whole
  destination rather than degrading gracefully. Validated and wrapped end to
  end now.
• A schedule with zero usable days (empty array, every day unparseable) was
  treated identically to a confirmed closure. Now distinguished from a
  schedule that explicitly marks every day Closed (a real signal — e.g.
  Volcano Bay's off-season) — only the former fails open (ungated).
• buildSchedules assumed OpenTimeString/CloseTimeString are always
  present; real UOR data omits them on some days. Fixed the same way (skip
  rather than publish an Invalid Date).
• Added missing coverage: Universal Orlando's 4-park case (previously
  untested — only single-park Hollywood was exercised), the venue
  reparenting branch end to end, the unmapped-venue fail-open branch,
  multi-day schedule scanning, and non-array/empty-array/all-closed
  response shapes.

Known residual risk — not fixed, flagged for follow-up

The type declared a SpecialEntryString field that isn't real; the actual
field is SpecialEntryUnix, present on every real day captured but 0 in
every sample checked across ~11 weeks (both resorts), including deep into a
would-be Halloween Horror Nights window. Its real meaning is unconfirmed —
possibly unused by this endpoint entirely, possibly only populated close to
an actual special-hours date. Left deliberately unread rather than guessing
at undocumented semantics.

The open question this leaves: does the legacy /venues/{id}/hours
endpoint this fix relies on cover ticketed after-hours events (e.g. HHN) at
all? No HHN-branded shows exist in the feed yet (event is weeks out), so
this can't be resolved from current data. If such an event's hours aren't
covered by this endpoint, shows tied to it would be wrongly gated CLOSED
during the live event — the opposite failure mode from the one this PR
fixes. Worth a live check once such an event is actually running; will file
a follow-up card for that.

Verification

• Full npm test suite green (2066 tests).
• npx tsc --noEmit clean.
• Live-verified against both US resorts while genuinely closed (~00:26 PDT /
  03:26 EDT, then re-verified after the hardening pass): Universal Studios
  Hollywood's 45 live rows and Universal Orlando's 96 live rows across all
  four parks are all CLOSED.

Test plan

• npm test — full suite passes
• npm run build — typechecks clean
• npm run dev -- universalstudios and npm run dev -- universalorlando against live upstream while both resorts are closed
• Three independent reviews (correctness, regression risk, test coverage) — findings addressed above
• Cube review

🤖 Generated with Claude Code

[cubehouse]

Holding this in draft. Found a real, confirmed risk after opening this PR: Halloween Horror Nights runs at both Universal Orlando and Universal Studios Hollywood as a wholly separate ticketed admission (day ticket ≠ HHN ticket, either direction), and the legacy /venues/{id}/hours schedule endpoint this fix clock-gates against shows zero signal of that evening window — confirmed against real data for 2026-08-28, Orlando's actual first HHN night this year (already locked into the schedule, only 9 days out at time of writing).

Unresolved: whether HHN-branded shows appear in show-list.json under the same day-park venue_id this fix gates. If they do, this fix would wrongly force them CLOSED during the live event.

Verifying live once HHN opens (~2026-08-28) rather than merging on a guess. Tracked as a gate on the internal board; will update here once resolved.