[MLton] adding synchronous finalizers

[Stephen Weeks]

> As to polling and signals, given that one would only do it (I think)
> in an otherwise single-threaded program, is there any point to the
> extra power of signal handlers have in the threaded case?

No, one probably one wouldn't use the extra power.  The point is that
the extra power is already there for multi-threaded programs.  So,
should we include checks to make sure it isn't used if one wants
synchronous finalizers?

Another problem -- one might want to treat finalizers synchronously
but handle signals asynchronously.  Don't ask me why, but they seem
like different enough beasts that it's possible.