fix hang with gdk-pixbuf >=2.44 Glycin: don't set SIGCHLD to IGNORE

[yennor]

Disclaimer: I'm not a Perl developper. I did the whole thing using a LLM. It looks sound to me, also passed tests and everything. But probably a seasoned perl developer should take a look at it @squentin ;-)

gdk-pixbuf 2.44 introduced Glycin, which decodes images in sandboxed child processes (bwrap). Setting $SIG{CHLD}='IGNORE' caused the kernel to auto-reap these children before GLib's GSubprocess could collect their exit status, making every gdk-pixbuf image load hang indefinitely.

Fix by removing $SIG{CHLD}='IGNORE' and instead properly reaping child processes at each fork site:

• forksystem and SaveTags: use Glib::child_watch_add
• metadata readers and simple_http_wget: use waitpid after completion
• mpv/mplayer _Kill_timeout: use waitpid(WNOHANG) to reap zombies

Also fix a pre-existing bug in forksystem where fork failure (undef) fell through into the child branch because undef==0 in numeric context.

Fixes: #262

[davidfoerster]

Which GLib bindings are you using? Mine don't feature Glib::child_watch_add but they do have Glib::Child->watch_add albeit with a different argument order.

[yennor]

You are completely right. As I said above. I didn't really do those changes myself. I just tested them, and told the model to test them, but as it seems, we never entered that code path... I know sooner or later I'll have to change my music player, so I didn't really invest too much time into it ;-)