tests/wait-p: Don't depend on yash-specific behavior

Previously, the test case 'trap interrupts wait' assumed that the
asynchronous subshell ran as a direct child process of the main shell.
This assumption was not POSIXly-portable. The shell is allowed to run
the subshell in a child process of the asynchronous job's process, in
which case the job process might be terminated by SIGUSR2 before the
trap action is performed in the subshell.

This commit rewrites the test so that the test result does not depend on
the wait status for the subshell.

tests/wait-p.tst

@@ -101,23 +101,19 @@ __IN__
 test_oE 'trap interrupts wait' -m
 interrupted=false
 trap 'interrupted=true' USR1
-(
-    set +m
-    trap 'echo received USR2; exit' USR2
-    while kill -s USR1 $$; do sleep 1; done # loop until signaled
-)&
-wait $!
+while kill -s 0 $$; do kill -s USR1 $$; done&
+# The asynchronous job should eventually interrupt the wait.
+wait
 status=$?
 echo interrupted=$interrupted $((status > 128))
 kill -l $status
-# Now, the background job should be still running.
+# Now the job should be still running. Kill it.
 kill -s USR2 %
-wait $!
+wait
 echo waited $?
 __IN__
 interrupted=true 1
 USR1
-received USR2
 waited 0
 __OUT__