feat(matrix): allow pkg.yml to opt in to windows/x86-64

[tannevaled]

Today get-matrix.ts throws "invalid platform" the moment a pkg.yml lists windows/ anywhere. That's a hard ceiling on Windows enablement even at the discussion level — there's no way to materialise the gap in actual CI.

Open the door without changing default behaviour:

• Extend the per-platform regex to accept (linux|darwin|windows). The default-platforms list (used when a pkg.yml omits platforms:) stays linux+darwin only. Existing packages get an identical matrix.
• Add a case 'windows/x86-64' to get_matrix() that targets the hosted windows-latest runner.
• Accept the shorthand windows (resolves to windows/x86-64; no arm64 Windows runners on GitHub Actions today).

This is intentionally a no-op for every package currently in the pantry — only a package that explicitly lists platforms: [windows/x86-64] is affected. Such a package will fail at brewkit/build@v1 (which has no Windows codepath today), which is the point: it surfaces brewkit's gap as a real CI failure rather than a planning-time throw, and gives a place to discuss Windows enablement with concrete build logs.

Why now

In pkgxdev/pkgm#87 I'm porting pkgm to Windows. The pkgm-side work is essentially done (per-user install_prefix to %LOCALAPPDATA%\pkgm, hardlink fallback for symlink_with_overwrite, .cmd stub wrappers, PATH_SEP awareness throughout).

Per jhheider's reply on pkgxdev/pkgx#607, the v2 dist layout already has 15 toolchain-layer packages built for Windows (bun.sh, cmake.org, curl.se, deno.land, git-scm.org, go.dev, libarchive.org, nasm.us, ninja-build.org, openssl.org, perl.org, python.org, rust-lang.org, sqlite.org, zlib.net) — but the manifest pipeline that produced them was rolled back about a year ago. So pkgm install <app> on Windows can't end-to-end work today without that pipeline coming back online. This PR takes the next-smallest enabling step on the path: it lifts the planning-time veto on windows/x86-64 so the brewkit-side conversation can happen with a concrete CI failure to point at, rather than getting cut short at "invalid platform".

This PR doesn't fix anything end-to-end. It removes the planning-level veto so brewkit/manifests work has a stable target.

Refs

• platform support megaticket pkgx#607 — platform-support megaticket and the rolled-back-manifests context (per maintainer reply).
• feat(windows): scaffolding pass for Windows support pkgm#87 — pkgm port to Windows (scaffolding pass, draft).
• libpkgx#48 + pandoc needs ffi #49 — Windows support in libpkgx itself, merged Oct 2023 with one known FFI-lockfile caveat.