More protection against SIGCHLD brokenness

[Bruce Korb]

> Oh, I was thinking more along the lines of:
> 
>   initialize_signals ()
> 
> or something at the beginning of main.
> 
> Either way.

We have three ways now :-)

1.  Just fix the executables that we know have the issue.
    That's the current patch.

2.  Create a main() procedure wrapper in libiberty to solve
    the problem for all programs that use libiberty for all time.

3.  Add a routine to libiberty that makes it a little easier
    to ensure sane signal handling for programs that call that
    routine.

Um, which "Either way"?  :-)  My preferences are this order,
with #2 second only because I dislike putting main in a library.