benchmark: add runtime funcinfo scaling probes

[cpunion]

Depends on #2010.

Adds reusable scaling probes to benchmark/runtime_funcinfo so runtime funcinfo performance can be compared against Go across project sizes and cold first-use paths.

Changes

• Add a cold scenario that measures first-use FuncForPC, Func.FileLine, runtime.Caller, and CallersFrames from a fresh process.
• Add -scales=PxM,... to generate multiple multipkg_* and cold_* scenarios in one run.
• Store scenario metadata in results.json, including package/function counts and target counts.
• Document the new cold and scale semantics.

Validation

• go test ./benchmark/runtime_funcinfo
• Go smoke: go run ./benchmark/runtime_funcinfo -out /tmp/llgo-runtime-funcinfo-scale-smoke -runs=1 -iters=1000 -scenarios=multipkg,cold -scales=2x2,3x2 -variant go=go
• LLGo smoke: go run ./benchmark/runtime_funcinfo -out /tmp/llgo-runtime-funcinfo-scale-llgo-smoke2 -runs=1 -iters=1000 -scenarios=cold -scales=2x2,3x2 -variant llgo=llgo,/tmp/llgo-scaling-bench,/Users/lijie/source/goplus/llgo-wt-runtime-funcinfo-scaling
• macOS arm64 scaling run with Go 1.26.4 and runtime: speed up funcinfo hot paths #2010: -runs=5 -iters=200000 -scenarios=multipkg,cold -scales=6x6,12x12,24x24

Cells are best/trimmed avg. multipkg.*Many and cold.Warm*Many are batch metrics over the target count in each row.

Go 1.26.4 vs #2010 Scaling Snapshot

scenario	targets	metric	Go 1.26.4	#2010
multipkg_6x6	36	FuncForPCMany	605/630ns	238/242.3ns
multipkg_6x6	36	FileLineMany	2/2.1us	312/319.3ns
multipkg_12x12	144	FuncForPCMany	2.5/2.5us	950/987ns
multipkg_12x12	144	FileLineMany	8/8.1us	1.3/1.3us
multipkg_24x24	576	FuncForPCMany	10.3/10.7us	3.9/3.9us
multipkg_24x24	576	FileLineMany	31.7/32.1us	5.1/5.1us
cold_6x6	36	FirstFuncForPC	2/2.2us	13.8/14ms
cold_12x12	144	FirstFuncForPC	2/2.3us	14.1/14.4ms
cold_24x24	576	FirstFuncForPC	2.5/3us	13.7/14.1ms
cold_24x24	576	WarmFuncForPCMany	10/10us	3.5/3.7us
cold_24x24	576	WarmFileLineMany	31.5/31.7us	4.9/4.9us

Interpretation

#2010 is faster than Go 1.26.4 on hot many-PC entry lookup in these generated probes, but still has a large cold first-use gap. The cold FirstFuncForPC results show lazy runtime funcinfo table initialization around 14 ms on macOS for this probe, while Go's linker-generated pclntab/findfunctab path stays in low microseconds.

[codecov]

Codecov Report

❌ Patch coverage is 51.36218% with 607 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
benchmark/runtime_funcinfo/main.go	0.00%	595 Missing ⚠️
cl/instr.go	97.51%	8 Missing and 4 partials ⚠️

📢 Thoughts on this report? Let us know!

[cpunion]

Superseded: the scaling harness was carried into #2016 as benchmark/runtime_funcinfo, now also covering ordinary-code (plain), call-depth and big-method scale dimensions.