ci: e2e (windows) job is chronically flaky under load (different specs fail each run) #189

The `e2e (windows)` CI job has failed with zero actual test failures on two consecutive runs of the same commit (PR #161, run 34401629220): all failures are Playwright "flaky" (fail once/twice, pass on retry), with a different set of unrelated specs each time:

• Run 1: `open-file-expand-folder`, `file-browser-sidebar-tab-switch`, `review-line-comment`, `session-timeline-lifecycle-state`, `session-timeline-projection`, `session-timeline-reasoning-projection`, `session-timeline-tool-projection`, `session-timeline-tool-state`, `terminal-composer-focus`, `smoke/session-timeline` — 9 flaky, 88 passed
• Run 2 (rerun of the same commit): `file-browser-sidebar-tab-switch`, `open-file-expand-folder`, `review-line-comment`, plus others — 11 flaky, 86 passed

No spec repeats as a hard failure across runs; different specs each time, spread across completely unrelated feature areas (file browser, review pane, session timeline, terminal). This is the signature of the Windows e2e runner being resource-starved (memory/CPU contention slowing down Playwright's own internal waits past their thresholds), not a real regression in any of them.

Playwright's default behavior treats any "flaky" (needed-a-retry) test as a non-zero exit code for the whole `test:e2e:local` script, so the CI check goes red even when 0 tests hard-fail.

Not investigated further — flagging so this stops looking like "the PR broke something" on every red `e2e (windows)` check. Possible directions: bump the Windows e2e runner tier/resources, increase Playwright's default expect/action timeouts for Windows specifically, or treat flaky-but-passed as non-fatal in CI (`--reporter` / retry accounting change).