fix(e2e): sweep stale 2026-06-07 dates + hardcoded solver windows

[tomqwu]

Why

Follow-up from #227. That PR bumped 2026-06-07 in the four e2e files that watch /v/open (which filters out past events), but explicitly deferred the remaining seven because their tests also seed a hardcoded solver window (from=2026-05-19, to=2026-06-30) that has to move in lockstep with the seed event date — bumping just one would push the seed outside the window and drop the assignment.

Today (2026-07-13) is already ~5 weeks past 2026-06-07. The seven files continue to compile and run, but any downstream assertion that depends on a solve-and-publish producing a real assignment (/v/schedule seeing "Sunday 10am Service", /v/inbox seeing the auto notification, the ICS feed containing the shift, the onboarding wizard hitting 2 of 4 done) is at real risk of quietly going empty as the seed date recedes and the fixed window falls out of relevance.

This PR closes the sweep in one commit.

What changed

• tests/e2e/_helpers.py — new solver_window_around(seed_iso) helper that returns a (from_date, to_date) pair three weeks either side of the seed. Same rot-prevention pattern as the existing next_sunday_iso().
• Inlined seed + window in single-test files (each was already calling page.fill("#ev_date", ...) and page.fill("#from_date"|"#to_date", ...) inside a single test function):
  • test_smoke_full_loop.py
  • test_availability_solver.py — same seed date used for /v/availability time-off entries as well
  • test_bulk_notify.py
  • test_calendar_subscription.py
  • test_onboarding_wizard.py
• Module-level constants for two files whose seed + window are shared across multiple test functions via helpers (_new_event, _solve_and_publish):
  • test_notif_assign_unpublish.py
  • test_analytics_recurring_swapreview.py
  • Chose EV_DATE / FROM_DATE / TO_DATE at import time so the helpers stay parameterless. next_sunday_iso() is pure w.r.t. date.today(), so a single test run sees one consistent triple.
• tests/unit/test_e2e_helpers.py — new: 3 invariants for the two helpers (Sunday-in-future, seed strictly inside window, deterministic given the same seed).

Validation

• poetry run pytest tests/unit/test_e2e_helpers.py tests/unit/test_e2e_lane.py → 5 passed
• make test-unit-fast → 341 passed, 21 skipped
• poetry run pytest tests/api tests/contract tests/web → 550 passed
• poetry run black --check api tests → clean
• poetry run ruff check api tests → clean

E2E lane could not be exercised locally (Playwright browser channel mismatch, same limitation noted in #227); CI is the source of truth for the e2e job. Contract snapshot untouched — no endpoints changed.

Follow-ups

• 2026-07-05 in test_analytics_recurring_swapreview.py (#rs_sd, a recurring-series start date on a different code path) is not on any solve-and-publish flow and is still ~2 weeks in the future. Left alone to keep this sweep narrow; worth bumping alongside the next rot fix.

---

Generated by Claude Code