feat(runner): distinguish instrumented command and tool error in codspeed

[not-matthias]

Added a small wrapper script, which executes cargo codspeed run in a subprocess and then writes the status code to a file.

Logs when the benchmark failed (in this case with a panic):

[DEBUG::codspeed::run::runner::valgrind::measure] Valgrind exit code = Some(0), Program exit code = 1
Error: failed to execute the benchmark process

Logs when valgrind error is detected (only an example, this doesn't cause valgrind to crash)

[DEBUG::codspeed::run::runner::valgrind::measure] valgrind.log: ==486607== Using source line as position.
--486607-- warning: L3 cache found, using its data for the LL simulation.
--486607-- warning: specified LL cache: line_size 64  assoc 12  total_size 18,874,368
--486607-- warning: simulated LL cache: line_size 64  assoc 18  total_size 18,874,368
==486607== 
==486607== Process terminating with default action of signal 4 (SIGILL): dumping core
==486607==  Illegal opcode at address 0x1205CF
==486607==    at 0x1205CF: my_bench::fibonacci (in /app/rust-crate/my_bench)
==486607==    by 0x120E57: codspeed_divan_compat::compat_divan::compat::bench::args::bench (in /app/rust-crate/my_bench)
==486607==    by 0x12313B: codspeed_divan_compat::compat_divan::compat::main (in /app/rust-crate/my_bench)
==486607==    by 0x121362: std::sys::backtrace::__rust_begin_short_backtrace (in /app/rust-crate/my_bench)
==486607==    by 0x121358: std::rt::lang_start::{{closure}} (in /app/rust-crate/my_bench)
==486607==    by 0x140DDF: std::rt::lang_start_internal (in /app/rust-crate/my_bench)
==486607==    by 0x12067B: main (in /app/rust-crate/my_bench)

Error: failed to execute valgrind

[not-matthias]

Depends on CodSpeedHQ/codspeed-rust#86

The correct exit code is now displayed (illegal instruction = signal 4)

[DEBUG::codspeed::run::runner::valgrind::measure] Valgrind exit code = Some(0), Valgrind signal = None, Program exit code = 132
Error: failed to execute the benchmark process, exit code: 132

[art049]

Is there an easy way for us to write a test of this behavior? At least testing the script isolated from the rest

[art049]

I'm not super confident merging this right now, mostly because of the quoting and wrapping of parameters.

In

cmd.args(["sh", "-c", &run_cmd]);

I think we should replace sh -c with our wrapper script directly. I think it would avoid messing with the quotes and formatting command as strings.

[art049]

just one last thing to simplify the script and I think we're good to go