docs: add contributor guide and validation wrappers

[cpunion]

Summary

• add a standard CONTRIBUTING.md for human contributors and link it from the README
• keep CLAUDE.md focused on automation-specific scope, worktree, upstream-write, and reporting safeguards; AGENTS.md continues to link to it
• centralize shared development, testing, coverage, IR, Go compatibility, cross-platform, Wasm, and embedded guidance in the human-facing contribution guide
• add dev/test_wasm.sh and dev/test_embed.sh for end-to-end target validation on macOS/Linux and Windows through WSL2
• document Docker Desktop/OrbStack, Rosetta, Docker/QEMU, CI platform coverage, and .github/codecov.yml exclusions
• make source-embedded LIT checks and litgen the sole documented IR-check workflow

Validation

• shellcheck -x dev/test_wasm.sh dev/test_embed.sh
• bash -n dev/test_wasm.sh dev/test_embed.sh
• ./dev/test_wasm.sh
• ./dev/test_embed.sh
• all-target empty fixture build: 186 successful, 0 failed; existing ignored/warning targets unchanged
• verified README.md and CLAUDE.md references to CONTRIBUTING.md, referenced anchors, and AGENTS.md -> CLAUDE.md
• git diff --check upstream/main...HEAD

No compiler or runtime code changed.

[fennoai]

Review summary

Documentation-only change: expands CLAUDE.md with contributor/validation guidance (testing, Go compatibility, cross-platform/Wasm/embedded validation, GitHub safety) and adds AGENTS.md as a symlink to CLAUDE.md.

I verified the added content against the repository. It is highly accurate: all referenced scripts, workflows, and files exist (dev/*.sh, chore/litgen, cl/_testrt/litdemo/in.go, embedded/ESP scripts, workflow files); command invocation forms match how the scripts actually parse arguments (dev/docker.sh <arch> [cmd...], dev/test_goroot.sh -- <runner-args>, litgen <file>); anchor links resolve (dev/README.md#6-refresh-test-goldens, README #dependencies/#how-to-install); and version claims match CI (LLVM 19, pinned Go patch releases). The // LITTEST marker convention is real, and AGENTS.md is a genuine 120000 symlink resolving to CLAUDE.md, consistent with the "links to this file" wording.

A few minor documentation-quality points are left inline. No blocking issues.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

LLGo baseline benchmarks

c370aebee583 | workflow run | long-term charts

Program measurements

Platform	Workload	File size	vs base	Build	vs base	Run	vs base
Linux	cprintf	18456 B	+0.0%	290.765 ms	+1.1% (worse)	1.343 ms	+2.7% (worse)
Linux	fmtprintf	1829968 B	+0.0%	3.071 s	+1.2% (worse)	3.143 ms	-1.8% (better)
Linux	println	68008 B	+0.0%	297.685 ms	+1.1% (worse)	1.635 ms	-0.9% (better)
macOS	cprintf	84672 B	+0.0%	423.707 ms	+15.3% (worse)	3.906 ms	+23.8% (worse)
macOS	fmtprintf	1869328 B	+0.0%	3.078 s	-37.7% (better)	13.993 ms	-29.7% (better)
macOS	println	121200 B	+0.0%	385.758 ms	+11.3% (worse)	4.263 ms	+5.9% (worse)

Core language and compiler benchmarks

Platform	Benchmark	ns/op	vs base
Linux	BenchmarkLookupPCRandom	13.310 ns/op	+0.3% (worse)
Linux	BenchmarkMergeCompilerFlags	152.600 ns/op	+0.8% (worse)
Linux	BenchmarkMergeLinkerFlags	94.740 ns/op	-2.6% (better)
Linux	BenchmarkChannelBuffered	33.680 ns/op	-0.1% (better)
Linux	BenchmarkChannelHandoff	27666 ns/op	-0.0% (better)
Linux	BenchmarkDefer	49.860 ns/op	-3.4% (better)
Linux	BenchmarkDirectCall	1.557 ns/op	+0.1% (worse)
Linux	BenchmarkGlobalRead	1.558 ns/op	+0.1% (worse)
Linux	BenchmarkGlobalWrite	2.482 ns/op	+0.1% (worse)
Linux	BenchmarkGoroutine	32177 ns/op	-1.9% (better)
Linux	BenchmarkInterfaceCall	8.099 ns/op	-0.0% (better)
Linux	BenchmarkRuntimeGetG	2.180 ns/op	-0.1% (better)
macOS	BenchmarkLookupPCRandom	12.140 ns/op	-5.5% (better)
macOS	BenchmarkMergeCompilerFlags	163.500 ns/op	-13.3% (better)
macOS	BenchmarkMergeLinkerFlags	85.420 ns/op	-14.8% (better)
macOS	BenchmarkChannelBuffered	22.090 ns/op	-17.3% (better)
macOS	BenchmarkChannelHandoff	7306 ns/op	-12.3% (better)
macOS	BenchmarkDefer	27.190 ns/op	-29.5% (better)
macOS	BenchmarkDirectCall	1.117 ns/op	-4.0% (better)
macOS	BenchmarkGlobalRead	1.018 ns/op	-16.6% (better)
macOS	BenchmarkGlobalWrite	1.117 ns/op	-14.6% (better)
macOS	BenchmarkGoroutine	28273 ns/op	-44.2% (better)
macOS	BenchmarkInterfaceCall	4.486 ns/op	-12.1% (better)
macOS	BenchmarkRuntimeGetG	2.069 ns/op	-12.0% (better)

Compared with d37108c85377 measured in the same runner job.