[Based on #106] debug: validate embedded GDB Remote artifacts

[cpunion]

Part of xgo-dev#2164 Task 4.

Embedded target configuration already records preferred GDB commands, but that information was not exposed to debugger launchers, and the final Cortex-M artifact had no end-to-end source-debugging acceptance test.

This PR:

• propagates target GDB candidates through crosscompile.Export;
• builds a real Cortex-M host ELF with DWARF and proves that stripping its debug sections leaves identical target-loadable bytes;
• validates the final artifact with both GDB Remote and LLDB gdb-remote against QEMU, including source breakpoints, locals, composite values, globals, and backtraces;
• documents the equivalent OpenOCD transport contract and runs the check in Targets CI.

LLVM 19 LLD can leave tombstone DIEs after section GC, so verification uses a temporary garbage-collected copy while both debuggers consume the original ELF. This avoids changing the artifact or dropping live global-variable DIEs.

Local verification:

• macOS: go test -timeout 20m ./internal/build -count=1
• macOS: go test -coverprofile=... ./internal/crosscompile — 78.6%
• Linux/ARM64: embedded QEMU + GDB Remote + LLDB remote integration
• Linux/ARM64: go test -coverprofile=... ./internal/crosscompile — 77.3%
• shellcheck and git diff --check

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[github-actions]

LLGo baseline benchmarks

a5d810975080 | workflow run | long-term charts

Program measurements

Platform	Workload	File size	vs main	Build	vs main	Run	vs main
Linux	cprintf	19344 B	+4.3% (worse)	310.521 ms	-24.7% (better)	1.069 ms	-31.1% (better)
Linux	fmtprintf	2218824 B	+0.2% (worse)	2.694 s	-30.1% (better)	2.145 ms	-34.7% (better)
Linux	println	73832 B	+1.6% (worse)	306.710 ms	-23.3% (better)	1.362 ms	-18.6% (better)
macOS	cprintf	84672 B	+0.0%	333.024 ms	-19.6% (better)	2.354 ms	-34.4% (better)
macOS	fmtprintf	2361696 B	+0.7% (worse)	4.499 s	+14.2% (worse)	26.606 ms	+13.8% (worse)
macOS	println	125888 B	-0.9% (better)	420.962 ms	+1.5% (worse)	4.670 ms	-35.7% (better)

Core language and compiler benchmarks

Platform	Benchmark	ns/op	vs main
Linux	BenchmarkLookupPCRandom	9.662 ns/op	-28.3% (better)
Linux	BenchmarkMergeCompilerFlags	112.100 ns/op	-27.4% (better)
Linux	BenchmarkMergeLinkerFlags	72.430 ns/op	-28.5% (better)
Linux	BenchmarkChannelBuffered	29.730 ns/op	-14.5% (better)
Linux	BenchmarkChannelHandoff	19485 ns/op	-39.2% (better)
Linux	BenchmarkDefer	35.190 ns/op	-37.5% (better)
Linux	BenchmarkDirectCall	1.364 ns/op	-12.7% (better)
Linux	BenchmarkGlobalRead	1.384 ns/op	-11.3% (better)
Linux	BenchmarkGlobalWrite	2.179 ns/op	-12.6% (better)
Linux	BenchmarkGoroutine	24224 ns/op	-68.0% (better)
Linux	BenchmarkInterfaceCall	6.818 ns/op	-12.5% (better)
Linux	BenchmarkRuntimeGetG	1.641 ns/op	-69.1% (better)
macOS	BenchmarkLookupPCRandom	11.540 ns/op	-7.2% (better)
macOS	BenchmarkMergeCompilerFlags	117 ns/op	-6.0% (better)
macOS	BenchmarkMergeLinkerFlags	62.200 ns/op	-10.3% (better)
macOS	BenchmarkChannelBuffered	20.500 ns/op	-26.1% (better)
macOS	BenchmarkChannelHandoff	6543 ns/op	-6.6% (better)
macOS	BenchmarkDefer	24.960 ns/op	-33.3% (better)
macOS	BenchmarkDirectCall	0.941 ns/op	-35.2% (better)
macOS	BenchmarkGlobalRead	0.942 ns/op	-22.2% (better)
macOS	BenchmarkGlobalWrite	0.946 ns/op	-36.4% (better)
macOS	BenchmarkGoroutine	56197 ns/op	+77.5% (worse)
macOS	BenchmarkInterfaceCall	4.100 ns/op	-24.0% (better)
macOS	BenchmarkRuntimeGetG	2.045 ns/op	-20.6% (better)

Compared only with the latest matching platform in the main series.

Warning

• Persistent publishing is unavailable because no data token is configured.