Carry the rename into the data layer

[isc]

Stacked on #224 — base branch is rename-to-arabesque, so review that one first.

#224 deliberately stopped at the surface, leaving the data under the old name.
This finishes the job without asking anyone to re-import a backup.

What moves

• IndexedDB — piano-trainer → arabesque. openDatabase() copies the old
  database's stores into the new one on first open, then deletes it.
• localStorage — the pt- and pt: prefixes (two separators, for no
  reason) collapse into a single arabesque:. legacyKeys.js renames the keys
  in place, including the per-score strictBpm:<url> ones.

Three details worth a look

• The legacy database is probed with indexedDB.databases(), not open():
  open() would create the very database it is testing for, making every future
  first-time visitor pay for a database they never had.
• The copy only runs into a database we just created (oldVersion === 0), so if
  a browser somehow has data under both names, the new one wins.
• legacyKeys.js runs as an import side effect, imported first by
  i18n.js, because i18n.js reads the stored language while its module body
  evaluates. Anything called from a page's entry script would arrive after that
  read and lose the language on the first load. For the same reason the landing
  page's inline redirect script — which runs before any module — reads both the
  new and the old key.

Test fix

test_helper.rb gains wait_for_store. The seeding tests opened the database
from Ruby after a fixed sleep, assuming the page had already created it. That
assumption was always fragile and the extra await here broke it: opening the
database first creates an empty one at the same version, which then never
upgrades, so the stores are missing for good and the library renders nothing.
They now wait on that condition rather than on the clock.

Verification

• Vitest: 97 passed.
• rake test: 66 runs, 1 failure — test_playback_starts_and_stops, which
  waits on piano samples fetched from a CDN. It is pre-existing and
  order-dependent: the parent branch fails the same test identically when run
  with the same seed (46255), and it passes in isolation on both. Worth fixing
  separately.

🤖 Generated with Claude Code