Incorrect use of Pdeathsig for killing child on linux.

[codefromthecrypt]

Describe the bug
Right now, on Linux, we set SysProcAttr.Pdeathsig = SIGTERM with a comment that this will kill the child when the parent dies. This is incorrect usage to achieve the results. The correct way to use Pdeathsig would be to launch a goroutine, lock it to the OS thread, start the child in that goroutine, and don't exit the goroutine until the child exits.

While revisiting this, we should re-check any assumptions for example that SIGTERM is the appropriate signal to catch.

Personally, I think we should prefer behaviour mentioned in the relevant go issue on conflict as many signal related code have had problems here.

Additional context
See golang/go#27505 (comment) for an example
See rootless-containers/rootlesskit#65 for a discussion leading to this

[dio]

We haven't done the suggested solution specifically, since the experiment in https://gist.github.com/dio/85d51d0c199b7598cefa56cb95521537 shows that it only matters for many children processes. However, we have added tests for making sure when killing func-e on Linux, we propagate SIGKILL to envoy. See: #372 (#371 changed Pdeathsig config from SIGTERM to SIGKILL).

[codefromthecrypt]

works for me. thanks for the research and if something changes in the future (like somehow we start a million envoys), we can reference this!