Fix --wait-pid false-negative for cross-user processes and pidof docs

processExists used kill(pid, 0) and treated any error as "gone", but
that syscall returns EPERM when the target PID exists but belongs to
a user the caller can't signal (e.g. ntfy running as a regular user
waiting on a root-owned restic process). The wait loop then exited
immediately with "process with PID ... not running" even though the
process was very much alive.

Treat EPERM as "exists" and only ESRCH as "gone". The loop stays
correct: EPERM while the target is alive, ESRCH once it exits.

Also fix the pidof example in docs/subscribe/cli.md: plain `pidof X`
returns space-separated PIDs when multiple matches exist, which
breaks --wait-pid's single-int argument. Use `pidof -s` (single-shot)
so the example works even with multiple matching processes.

cmd/publish_unix.go

@@ -2,9 +2,15 @@
 
 package cmd
 
-import "syscall"
+import (
+	"errors"
+	"syscall"
+)
 
 func processExists(pid int) bool {
 	err := syscall.Kill(pid, syscall.Signal(0))
-	return err == nil
+	if err == nil {
+		return true
+	}
+	return errors.Is(err, syscall.EPERM)
 }

docs/subscribe/cli.md

@@ -111,7 +111,7 @@ Or, if you already started the long-running process and want to wait for it usin
 
 === "Using a `pidof`"
     ```
-    $ ntfy pub --wait-pid $(pidof rsync) mytopic | jq .
+    $ ntfy pub --wait-pid $(pidof -s rsync) mytopic | jq .
     {
       "id": "orM6hJKNYkWb",
       "time": 1655825827,