GHA free windows runners are 2vCPU instances, and are our longest-running jobs most of the time now.
Test using Warp's 4x windows runners to get a feel for runtime improvements and whether it might be worth migrating.
First run should save caches (even on a PR) so a re-run can test with caches also coming via warp.
Route the Windows native vcpkg caches through the existing internal cache wrappers so bitcoin/bitcoin can use WarpBuilds/cache on Warp Windows runners while forks keep the GitHub Actions cache path.
Also fix the runner expression quoting so the Windows Warp runner label is parsed as a string.
Allow bitcoin/bitcoin pull requests from branches in the main repository to save Windows vcpkg Warp caches so a trial PR can compare cold and warm rerun timings.
Keep fork pull requests restore-only to avoid letting untrusted PR code write to the shared cache namespace.
<!--e57a25ab6845829454e8d69fc972939a-->
The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.
<!--006a51241073e994b41acfe9ec718e94-->
For details see: https://corecheck.dev/bitcoin/bitcoin/pulls/35683.
<!--021abf342d371248e50ceaed478a90ca-->
See the guideline and AI policy for information on the review process. A summary of reviews will appear here.
<!--174a7506f384e20aa4161008e828411d-->
Reviewers, this pull request conflicts with the following ones:
If you consider this pull request important, please also help to review the conflicting pull requests. Ideally, start with the one that should be merged first.
<!--5faf32d7da4f0f540f40219e4f7537a3-->
Apparently Windows 2025 with VS2026 is coming in an image next week, so will look out for that and re-run when it's released.
We are using about 70-90k minutes per month on GH, if these runners were 25% faster we'd be looking at a ~$350 cost, and if they were 50% faster could be as low as ~$250/month.
8x runners cost double the 4x runners, so unless they're > 50% faster than the 4x, they'd cost us more.