[MLton] Main.main idiom

[Stephen Weeks]

> but I think it is better programming practice to not scatter
> top-level effects around.  It can actually be useful to compile and
> run the program without the call to Main.main -- it verifies that
> there weren't any unexpected evaluations in the program.

Fair enough.  This gives me another idea.  We could have a switch
(like -safe) that controls some boolean variable that is used to
govern whether or not the main function is called.  So, one would
write:

	val () = if MLton.runMain then Main.main () else ()

and then compile with -run-main true to generate the executable
(although maybe -run-main true would be the default).  One advantage I
see of this over -mainfn is that the programmer gets to specify within
the language how to call main.