fix startup_cmd SIGCHLD handler

[djpohly]

Looks like ignored signals aren't reset by exec() calls. If the startup command needs to fork and wait on child processes (e.g. a session manager like s6-svscan), the SIG_IGN that is still in place will interfere with waiting.

[sevz17]

Looks like ignored signals aren't reset by exec() calls.

So we could do this for spawn() too, right?

[djpohly]

Good call. If it makes the code simpler/easier to understand, I guess we could forgo the SIG_IGN altogether and just set a handler, which would then be reset by exec.

[sevz17]

How many lines would it add?

I am a bit reluctant to add more lines, in wlroots-next there are already 2230 lines.

[sevz17]

We could also entirely give up in the SLOC limit and try to make the code more readable/correct, but still trying to do things in the simplest way.

I am only concerned about SLOC because it's one of the principles of dwl, but I would not mind if we remove the limit.

[djpohly]

Oh, I'll watch my SLOC footprint for sure! I had a hunch the suggestion would actually reduce lines, and it turns out it did. This PR now reduces the count from cloc dwl.c by 5.

[sevz17]

Thank you!