fix(embedded-e2e): harden Playwright helpers against werkzeug transients

Add defensive retry/timeout wrappers for the dev-server flake patterns
observed in CI (socket hang up, ERR_EMPTY_RESPONSE, ECONNRESET, single-
threaded werkzeug starvation during SQL Lab queries).

- helpers/api/requests.ts: withTransientRetry wrapper retries
  page.request calls (CSRF + GET/POST/PUT/PATCH/DELETE) up to 3 times on
  transport-layer drops with 250ms linear backoff.
- helpers/navigation.ts (new): gotoWithRetry retries page.goto on
  ERR_EMPTY_RESPONSE / ERR_CONNECTION_RESET / socket hang up.
- pages/{Sql,DashboardList,ChartList,DatasetList,Dashboard,
  ChartCreation,CreateDataset}Page.ts: all goto() entry points now use
  gotoWithRetry.
- pages/SqlLabPage.ts: bump waitForPageLoad, executeQuery
  (waitForPost + .ant-steps visible) and waitForQueryResults timeouts
  from QUERY_EXECUTION (15s) to SLOW_TEST (60s); werkzeug round-trips
  routinely exceed 15s when cross-project workers contend for the
  single Python thread.
- components/core/EditableTabs.ts: addTab prefers the
  .ant-tabs-nav-operations dropdown trigger to dodge overflow flakes.
- components/modals/EditDatasetModal.ts: add waitForTextControlDebounce
  (350ms) after fillName / fillColumnDateFormat to settle the controlled
  input before save.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>