fix(update): avoid update resource while the process is zombie

[jokemanfire]

add a check in zombie init process.

If update task between the wait_pid call and init exit. The bug will occur, the cgroup maybe update the root's cgroup path.
Because the cgroup path has been recycled by system ,but the wait_pid haven't recycled the resource of the process. The update task is processing.

A easy way to reproduce this bug. Occasionally triggered in high concurrency scenarios.

1. start a shim ,like cri
   crictl run --no-pull container-config.json pod.json
2. gdb -p {target shim pod}
   (gdb) handle SIGCHLD ignore
   (gdb) continue
3. kill the init process

the init process will become zombie , you can send update task to the shim ,but the cgroup path is not correct.

[jokemanfire]

@mxpv @Mossaka Project CI may have some problems , could fix it ?

[mxpv]

@jokemanfire can you rebase the PR? CI should be fixed #395

[jokemanfire]

@mxpv fixed

[jokemanfire]

@mxpv it looks like IT fail ,I checked the logs, but there's no clue. Can you take a look？