windows: integrate native Windows platform support

[cpunion]

Part of xgo-dev#2325.

This draft is the single Windows integration and CI branch. It intentionally keeps the implementation together until native Windows support and its cross-platform regression coverage are complete. Reviewable upstream slices will be recreated only after this branch reaches that gate.

Integrated scope:

• MSVC target triples, COFF linking, architecture feature flags, and cross-compilation configuration;
• Windows C ABI lowering, FFI closures, system callbacks, and 386/amd64/arm64 calling-convention coverage;
• native threads, synchronization, TLS, GC registration, defer, panic/recover, SEH unwinding, signals, filesystem/process support, IOCP, sockets, and testing hooks;
• Windows-aware standard-library symbol coverage and repository test infrastructure;
• guards against runtime or binary-size overhead on non-Windows targets.

Validation completed so far:

• Windows 11 ARM64 local host/compiler, runtime, FFI, testing, protected-memory, and standard-library coverage checks;
• Windows AMD64 GitHub host smoke and standard-library coverage checks;
• Linux and macOS regression tests for shared runtime changes.

Completion gate for this integration PR:

• full test/std/..., test/goroot, the remaining test/... language and LLGo-extension suites, and applicable demos on Windows;
• Windows amd64, arm64, and 386 compile/runtime coverage where the toolchain supports execution;
• cross-compilation toolchain validation;
• review of runtime/standard-library diffs for Go source reuse and necessity;
• Linux/macOS size, performance, and test regression checks.

Windows tests are not skipped to make the branch green. Platform-specific failures must either be fixed or explicitly classified with a documented reason.

[github-actions]

LLGo baseline benchmarks

46fda6827fab | workflow run | long-term charts

Program measurements

Platform	Workload	File size	vs base	Build	vs base	Run	vs base
Linux	cprintf	19640 B	+352 B / +1.8% (worse)	345.697 ms	+14.41 ms / +4.4% (worse)	1.352 ms	-18 us / -1.3% (better)
Linux	cprintf-lto	19472 B	+352 B / +1.8% (worse)	338.435 ms	+5.971 ms / +1.8% (worse)	1.428 ms	+1.377 us / +0.1% (worse)
Linux	fmtprintf	1844080 B	+4992 B / +0.3% (worse)	2.688 s	-2.104 ms / -0.1% (better)	3.542 ms	-23.23 us / -0.7% (better)
Linux	fmtprintf-lto	1764016 B	+4544 B / +0.3% (worse)	13.551 s	-14.74 ms / -0.1% (better)	3.540 ms	-23.94 us / -0.7% (better)
Linux	println	69808 B	+1032 B / +1.5% (worse)	339.872 ms	+6.268 ms / +1.9% (worse)	1.683 ms	-34.94 us / -2.0% (better)
Linux	println-lto	63136 B	+672 B / +1.1% (worse)	554.304 ms	+5.503 ms / +1.0% (worse)	1.742 ms	+6.345 us / +0.4% (worse)
macOS	cprintf	84672 B	0 B / +0.0%	369.417 ms	+28.12 ms / +8.2% (worse)	2.345 ms	-13.17 us / -0.6% (better)
macOS	cprintf-lto	100912 B	0 B / +0.0%	350.507 ms	+39.51 ms / +12.7% (worse)	2.163 ms	-46.58 us / -2.1% (better)
macOS	fmtprintf	1877024 B	+544 B / +0.02899% (worse)	2.006 s	-166.5 ms / -7.7% (better)	11.498 ms	+416.3 us / +3.8% (worse)
macOS	fmtprintf-lto	1609920 B	+208 B / +0.01292% (worse)	13.969 s	+809.6 ms / +6.2% (worse)	6.080 ms	+1.402 ms / +30.0% (worse)
macOS	println	121840 B	+480 B / +0.4% (worse)	345.892 ms	+41.19 ms / +13.5% (worse)	3.534 ms	+222.3 us / +6.7% (worse)
macOS	println-lto	128688 B	+160 B / +0.1% (worse)	432.270 ms	+13.24 ms / +3.2% (worse)	2.658 ms	-89.46 us / -3.3% (better)

Core language and compiler benchmarks

Platform	Benchmark	ns/op	vs base
Linux	BenchmarkLookupPCRandom	12.220 ns/op	-0.03 ns/op / -0.2% (better)
Linux	BenchmarkMergeCompilerFlags	144.500 ns/op	-2.2 ns/op / -1.5% (better)
Linux	BenchmarkMergeLinkerFlags	93.910 ns/op	0 ns/op / +0.0%
Linux	BenchmarkChannelBuffered	36.350 ns/op	+0.05 ns/op / +0.1% (worse)
Linux	BenchmarkChannelHandoff	23614 ns/op	-1409 ns/op / -5.6% (better)
Linux	BenchmarkDefer	44.280 ns/op	-0.73 ns/op / -1.6% (better)
Linux	BenchmarkDirectCall	1.757 ns/op	-0.001 ns/op / -0.1% (better)
Linux	BenchmarkGlobalRead	1.757 ns/op	-0.001 ns/op / -0.1% (better)
Linux	BenchmarkGlobalWrite	2.808 ns/op	-0.001 ns/op / -0.0356% (better)
Linux	BenchmarkGoroutine	35399 ns/op	+5298 ns/op / +17.6% (worse)
Linux	BenchmarkInterfaceCall	8.929 ns/op	-0.176 ns/op / -1.9% (better)
Linux	BenchmarkRuntimeGetG	1.759 ns/op	-0.353 ns/op / -16.7% (better)
macOS	BenchmarkLookupPCRandom	13.210 ns/op	+2.56 ns/op / +24.0% (worse)
macOS	BenchmarkMergeCompilerFlags	142.400 ns/op	+41.7 ns/op / +41.4% (worse)
macOS	BenchmarkMergeLinkerFlags	86.630 ns/op	+20.22 ns/op / +30.4% (worse)
macOS	BenchmarkChannelBuffered	23.050 ns/op	+2.1 ns/op / +10.0% (worse)
macOS	BenchmarkChannelHandoff	8387 ns/op	+2113 ns/op / +33.7% (worse)
macOS	BenchmarkDefer	33.440 ns/op	+6.44 ns/op / +23.9% (worse)
macOS	BenchmarkDirectCall	1.112 ns/op	-0.014 ns/op / -1.2% (better)
macOS	BenchmarkGlobalRead	1.086 ns/op	+0.065 ns/op / +6.4% (worse)
macOS	BenchmarkGlobalWrite	1.043 ns/op	-0.075 ns/op / -6.7% (better)
macOS	BenchmarkGoroutine	55725 ns/op	+24526 ns/op / +78.6% (worse)
macOS	BenchmarkInterfaceCall	4.544 ns/op	-0.291 ns/op / -6.0% (better)
macOS	BenchmarkRuntimeGetG	2.129 ns/op	+0.168 ns/op / +8.6% (worse)

Compared with 0314faeba69d measured in the same runner job.