debug: allow supported targets to retain DWARF

[cpunion]

Part of xgo-dev#2164.

Fixed targets previously set AlwaysOmit and passed -S unconditionally, so -ldflags=-w=false could not produce a debug ELF. Once retention is enabled, direct ld.lld targets also must not receive the clang-only -gdwarf-4 link option.

This change:

• models DWARF retention and preserve/omit linker flags as target capabilities;
• lets supported ELF and Wasm linkers retain DWARF while preserving the current default omission policy;
• keeps clang-driver and direct-linker arguments separate;
• verifies a Cortex-M debug ELF contains DWARF and produces exactly the same flashed .bin bytes as its stripped control.

Validation:

• macOS: internal/crosscompile and relevant internal/build tests; full internal/build suite passed with the pre-existing hanging TestRunPrintfWithStdioNobuf excluded (76.2% package coverage, changed policy functions 100%);
• Linux/amd64: relevant internal/crosscompile and internal/build tests passed;
• end-to-end cortex-m-qemu build: explicit -w=false retains .debug_info, current default omits it.

[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

2d13fd37a1e2 | workflow run | long-term charts

Program measurements

Platform	Workload	File size	vs main	Build	vs main	Run	vs main
Linux	cprintf	18544 B	+0.0%	358.828 ms	-13.0% (better)	1.375 ms	-11.4% (better)
Linux	fmtprintf	2217608 B	+0.2% (worse)	3.156 s	-18.2% (better)	2.721 ms	-17.2% (better)
Linux	println	71504 B	-1.6% (better)	348.829 ms	-12.8% (better)	1.725 ms	+3.1% (worse)
macOS	cprintf	84672 B	+0.0%	308.578 ms	-25.5% (better)	2.328 ms	-35.1% (better)
macOS	fmtprintf	2361520 B	+0.7% (worse)	2.500 s	-36.5% (better)	16.514 ms	-29.3% (better)
macOS	println	125712 B	-1.0% (better)	305.466 ms	-26.3% (better)	3.201 ms	-55.9% (better)

Core language and compiler benchmarks

Platform	Benchmark	ns/op	vs main
Linux	BenchmarkLookupPCRandom	12.310 ns/op	-8.7% (better)
Linux	BenchmarkMergeCompilerFlags	144.900 ns/op	-6.2% (better)
Linux	BenchmarkMergeLinkerFlags	94.850 ns/op	-6.4% (better)
Linux	BenchmarkChannelBuffered	36.390 ns/op	+4.7% (worse)
Linux	BenchmarkChannelHandoff	25081 ns/op	-21.7% (better)
Linux	BenchmarkDefer	47.440 ns/op	-15.8% (better)
Linux	BenchmarkDirectCall	1.758 ns/op	+12.5% (worse)
Linux	BenchmarkGlobalRead	1.760 ns/op	+12.8% (worse)
Linux	BenchmarkGlobalWrite	2.806 ns/op	+12.5% (worse)
Linux	BenchmarkGoroutine	30399 ns/op	-59.8% (better)
Linux	BenchmarkInterfaceCall	8.794 ns/op	+12.8% (worse)
Linux	BenchmarkRuntimeGetG	1.760 ns/op	-66.8% (better)
macOS	BenchmarkLookupPCRandom	10.630 ns/op	-14.5% (better)
macOS	BenchmarkMergeCompilerFlags	95.970 ns/op	-22.9% (better)
macOS	BenchmarkMergeLinkerFlags	71.860 ns/op	+3.6% (worse)
macOS	BenchmarkChannelBuffered	23.150 ns/op	-16.5% (better)
macOS	BenchmarkChannelHandoff	7915 ns/op	+12.9% (worse)
macOS	BenchmarkDefer	29.540 ns/op	-21.1% (better)
macOS	BenchmarkDirectCall	1.084 ns/op	-25.4% (better)
macOS	BenchmarkGlobalRead	0.943 ns/op	-22.1% (better)
macOS	BenchmarkGlobalWrite	0.973 ns/op	-34.5% (better)
macOS	BenchmarkGoroutine	39159 ns/op	+23.7% (worse)
macOS	BenchmarkInterfaceCall	4.341 ns/op	-19.6% (better)
macOS	BenchmarkRuntimeGetG	2.312 ns/op	-10.3% (better)

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

Warning

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