DO NOT MERGE: Revert CLI e2e tests to check Windows E2E runtime

[gcsecsey]

Related issues

• Related to AINFRA-2588 (Windows E2E hangs in Buildkite)

How AI was used in this PR

Claude Code cross-checked the set of PRs that introduced the new CLI e2e harness, built the revert branch, and drafted this description.

Proposed Changes

⚠️ Do not merge. This PR is a CI experiment only.

The Windows E2E job started hanging to its 180-minute timeout around June 29. The investigation in AINFRA-2588 traced the onset to the wave of PRs that introduced the new real-CLI e2e test harness (apps/cli/commands/site/tests/helpers/cli-e2e.ts). Reverting only the Blueprint test (#3961, reverted in #4029) did not fix it — the next Windows run still timed out after ~3 hours. This PR reverts the entire harness so we can observe whether the Windows E2E job returns to its normal runtime without it.

• Reverts the remaining merged CLI e2e test PRs:
  • Add CLI integration test for site create (custom name, domain, HTTPS) #3947
  • Add CLI integration test for site start/stop lifecycle #3977
  • Add CLI e2e tests for site management (rename, PHP version, delete) #3989
• Add CLI integration test for creating a site from a custom Blueprint #3961 was already reverted on trunk via Revert CLI blueprint e2e test (#3961) and Windows E2E CI disable (#4008) #4029, so after this PR no trace of the harness remains.
• Temporarily re-enables the windows-x64 E2E matrix entry (reverts Disable Studio Windows E2E CI #4039) so this PR's build actually runs the Windows E2E suite. This commit must be dropped or reconciled with AINFRA-2588 before any merge.

Two caveats when reading the results: the new CLI e2e tests are vitest tests, so they never ran inside the Playwright E2E job itself — any effect on that job would be indirect (e.g. leaked process-manager-daemon.mjs processes or ~/.studio state accumulating on the persistent Windows agents, matching the CAPACITY_LIMIT_REACHED 36/36 errors seen in the logs). Also, #4034 and #4047 added new Playwright tests (existing apps/studio/e2e harness, not the CLI harness) after Windows E2E was disabled, so this PR's Windows run exercises those for the first time; they are left in place here.

Testing Instructions

1. Open the Buildkite build for this PR and watch the E2E Tests on windows-x64 step.
2. Compare its runtime against recent trunk Windows E2E runs (which hung to the 180-minute timeout, e.g. builds #18367/#18459).
3. A run that reaches a terminal state well under the timeout (comparable to the mac-arm64 step) supports the harness being the culprit; another 3-hour hang rules it out.

Pre-merge Checklist

• Have you checked for TypeScript, React or other console errors?