Comparing bpf helper probe results with `bpftool`

[rgo3]

With #375 we added support to probe the running kernel for helper availability.

As bpftool/libbpf also support feature probing, we have been manually comparing results. To reduce some effort I put together a small CLI program (bpftoolcmp) that shells out to bpftool and compares its results with results from the APIs we have in the lib.

It's output currently looks like the following:

❯ go run -exec sudo cmd/bpftoolcmp/main.go
Running bpftool -j feature ...

Comparing available program types ...

Comparing available helper functions ...
    False negative: API got different result than bpftool for: RawTracepoint/FnGetFuncIp
    False negative: API got different result than bpftool for: Syscall/FnSkStorageGet
    False negative: API got different result than bpftool for: Syscall/FnSkStorageDelete
    False negative: API got different result than bpftool for: Syscall/FnDPath
    False negative: API got different result than bpftool for: Syscall/FnGetFuncIp
    False negative: API got different result than bpftool for: TracePoint/FnGetFuncIp
    False negative: API got different result than bpftool for: PerfEvent/FnGetFuncIp
    False negative: API got different result than bpftool for: RawTracepointWritable/FnGetFuncIp

Note: Probes for the FnGetFuncIp helper are not actually a false negative, but a false positive on bpftools end: see here. The rest is still to be investigated.

Instead of such a CLI tool that would be run manually, @lmb suggested making it part of the testsuite of the features package. E.g. adding a test like TestBPFToolComapre(). I also would prefer having this as an automated test in the suite, however a test like this will put a burden on us to investigate the lib and bpftool/libbpf in case of a new failure in my opinion.