test: reduce repeated Windows pregate fixture setup

[atomantic]

What changed

The pregate hidden-content tests now build their committed Git fixture once and copy it for each isolated case. This removes repeated source copies and Git commits from Windows setup while preserving the real Git, CLI, index, and destructive-fixture assertions. Each copy refreshes its index metadata before the test snapshots it.

Earlier fixes already on main batch the Brain pagination seed (#8496), reuse the primary-checkout Git sandbox (#6003), and give the first detached process test its bounded Windows startup deadline (#8496). This PR finishes the remaining repeated pregate fixture setup in #8463. A full Windows CI run also exposed that deadExports.test.js assumed origin/main exists in a shallow workflow-dispatch checkout. The test now uses the prior commit as its comparison base when that ref is absent, while pull-request CI continues to use its exact CI_BASE_SHA.

Validation

• Focused pregate and dead-export suites: 24 tests passed; the named primary-checkout, Brain pagination, and detached-process suites also passed locally.
• npm run pregate on the final branch diff: 30 files, 405 tests passed.
• Local tool-free Codex review: no findings.
• Two consecutive full CI runs passed on commit 5424a0363: run 36207675569 and run 36208055736. All three Windows server shards and both CI gates succeeded in each run.
• Windows suite timings across those runs: pregate 23.3s / 16.8s; Brain pagination 15.8s / 15.3s; primary checkout guard 81.1s / 74.8s; detached spawn 44.0s / 35.8s. These are whole-suite durations; every setup hook completed within its existing limit.
• The first full run on the earlier head exposed the missing Git ref (run 36207310991), which the final-head runs verified fixed.

Closes #8463