[Based on #130] debug: add cross-platform session entry point

[cpunion]

Part of xgo-dev#2164 Task 4. Based on #130.

llgo lldb opens an existing artifact, while embedded debugging still required callers to build, start a server, connect, and clean up each process manually.

This change:

• adds llgo debug to build one executable package with DWARF and an -O0 debug default;
• routes native sessions to LLDB and non-Wasm embedded sessions to GDB, with explicit backend overrides;
• starts and cleans up target-configured debug servers or OpenOCD, and supports externally managed -remote sessions;
• keeps llgo lldb as the explicit existing-artifact command;
• moves the native LLDB and Cortex-M GDB/LLDB Remote integration checks through the new entry point.

The command reserves wasmtime and browser backend identities; their transports remain Tasks 5 and 6.

Local validation:

• macOS arm64: command packages, LLDB integration, race, and vet passed;
• Linux arm64: command packages, native GDB, native LLDB, and Cortex-M QEMU GDB/LLDB Remote passed;
• cmd/internal/debug coverage is 78.2% on both platforms.

[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

5fd8839902f4 | 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)	369.063 ms	-10.5% (better)	1.369 ms	-11.8% (better)
Linux	fmtprintf	2218832 B	+0.2% (worse)	3.517 s	-8.8% (better)	2.500 ms	-23.9% (better)
Linux	println	73832 B	+1.6% (worse)	372.590 ms	-6.9% (better)	1.616 ms	-3.4% (better)
macOS	cprintf	84672 B	+0.0%	340.278 ms	-17.9% (better)	2.331 ms	-35.0% (better)
macOS	fmtprintf	2361696 B	+0.7% (worse)	2.570 s	-34.8% (better)	16.415 ms	-29.8% (better)
macOS	println	125888 B	-0.9% (better)	313.988 ms	-24.3% (better)	3.390 ms	-53.3% (better)

Core language and compiler benchmarks

Platform	Benchmark	ns/op	vs main
Linux	BenchmarkLookupPCRandom	13.250 ns/op	-1.7% (better)
Linux	BenchmarkMergeCompilerFlags	159.900 ns/op	+3.5% (worse)
Linux	BenchmarkMergeLinkerFlags	105.100 ns/op	+3.8% (worse)
Linux	BenchmarkChannelBuffered	34.890 ns/op	+0.4% (worse)
Linux	BenchmarkChannelHandoff	28648 ns/op	-10.6% (better)
Linux	BenchmarkDefer	46.250 ns/op	-17.9% (better)
Linux	BenchmarkDirectCall	1.558 ns/op	-0.3% (better)
Linux	BenchmarkGlobalRead	1.556 ns/op	-0.3% (better)
Linux	BenchmarkGlobalWrite	2.482 ns/op	-0.5% (better)
Linux	BenchmarkGoroutine	33873 ns/op	-55.2% (better)
Linux	BenchmarkInterfaceCall	8.089 ns/op	+3.8% (worse)
Linux	BenchmarkRuntimeGetG	2.177 ns/op	-59.0% (better)
macOS	BenchmarkLookupPCRandom	10.680 ns/op	-14.1% (better)
macOS	BenchmarkMergeCompilerFlags	100.400 ns/op	-19.4% (better)
macOS	BenchmarkMergeLinkerFlags	59.330 ns/op	-14.4% (better)
macOS	BenchmarkChannelBuffered	21.140 ns/op	-23.8% (better)
macOS	BenchmarkChannelHandoff	6450 ns/op	-8.0% (better)
macOS	BenchmarkDefer	29.980 ns/op	-19.9% (better)
macOS	BenchmarkDirectCall	0.948 ns/op	-34.8% (better)
macOS	BenchmarkGlobalRead	0.943 ns/op	-22.1% (better)
macOS	BenchmarkGlobalWrite	0.941 ns/op	-36.7% (better)
macOS	BenchmarkGoroutine	20408 ns/op	-35.5% (better)
macOS	BenchmarkInterfaceCall	4.460 ns/op	-17.4% (better)
macOS	BenchmarkRuntimeGetG	1.882 ns/op	-27.0% (better)

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

Warning

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