feat(jq): add windows/x86-64 cross-compile pilot (#346 Phase 0)

[tannevaled]

Draft — Phase-0 pilot for pkgxdev/brewkit#346 (native Windows bottles RFC).

What

Extends the existing jq recipe to also build on windows/x86-64 (and trivially windows/aarch64), via cross-compile from Linux/macOS runners using the llvm.org/mingw-w64 toolchain in #12984.

End-to-end loop demonstrated

linux/x86-64 runner
       │
       ▼
build.script
       │  detects windows/* → switches CC + --host
       ▼
configure (host=x86_64-w64-mingw32, CC=x86_64-w64-mingw32-clang, --disable-onigjq)
       │
       ▼
make → make install
       │
       ▼
{{prefix}}/bin/jq.exe           ◄── cross-compiled PE/COFF binary
       │
test.script
       │  detects windows/*
       ▼
wine64 jq.exe .devs[1].github < test.json
       │
       ▼
assert == "jhheider"
       │
       ▼
PASS

This closes the Windows-port architectural loop sketched at pkgxdev/brewkit#346 with a concrete, reviewable recipe that actually produces + tests a Windows binary using only Linux CI.

Depends on

#12984	llvm.org/mingw-w64 toolchain
#12986	winehq.org headless wine

Both are draft PRs. This PR currently has soft fallbacks (PE magic-byte check when wine is missing) so it can be CI-validated even before #12986 lands — but the full runtime-validation story needs all three.

Known pilot limitations (called out in the recipe comments)

• Disables onigjq (regex) on Windows since oniguruma doesn't have a Windows bottle yet
• provides: bin/jq doesn't account for the .exe suffix on Windows — audit might complain; needs a brewkit hook
• Unconditional llvm.org/mingw-w64 build dep: harmless on non-Windows but ideally conditional once brewkit conditional-dep semantics are settled

What this is NOT

• Not a complete Windows port of jq (no regex on Windows)
• Not blocking the production linux/darwin jq build (current behaviour preserved)
• Not advocating for "Windows everywhere" — opt-in per recipe

Test plan

• linux/x86-64 CI: existing native build/test still passes
• linux/aarch64 CI: existing native build/test still passes
• darwin/* CI: existing native build/test still passes
• windows/x86-64 CI (once brewkit platform support lands): cross-compiles + runs through wine
• Manual: wine64 jq.exe . smoke-tests the resulting binary on a non-CI Linux box

Refs: pkgxdev/brewkit#346 (RFC), pkgxdev/brewkit#347 (fix-pe + bkwinvenv), pkgxdev/pkgx#607.

[tannevaled]

Note: CI fails by design until #12984 lands.

The CI failure is:

error: Uncaught (in promise) Error: pkg not found: llvm.org/mingw-w64

…because this PR adds llvm.org/mingw-w64 as a build dep, but that recipe is in #12984 which hasn't been merged yet. brewkit can't resolve the dep closure against pantry's main branch.

Two paths forward:

• Wait for new(llvm.org/mingw-w64): LLVM-based mingw-w64 cross-compiler (Windows-target toolchain) #12984 to land, then this PR's CI auto-greens
• Remove the llvm.org/mingw-w64 dep from this PR temporarily — but then the recipe can't actually cross-compile (just documents intent), which defeats the pilot's purpose

Sticking with path #1 — this PR stays draft + CI-red until the toolchain is available.

For non-Windows builds the recipe should still work (Windows path is gated by if [ "{{hw.platform}}" = "windows" ]). The dep-resolution-time failure happens for ALL platforms though, because brewkit checks deps before evaluating platform-conditionals. That's a brewkit-side limitation worth tracking — see open question on pkgxdev/brewkit#346:

Should brewkit gain conditional dependencies: { foo: '*', if: <predicate> } syntax (mirroring what already works on script steps)?

If yes, this PR can ship with llvm.org/mingw-w64: { '*', if: 'windows' } and CI greens immediately for linux/darwin builds.