[MLton] Finished (?) MLton.Child

[Stephen Weeks]

> I think it's ready from my side. I have attached two patches. 

Great.  I just checked in (a heavily modified form of) the patch.
Here's the commit email:

	http://mlton.org/pipermail/mlton/2004-November/016787.html

Here are the changes from the patch you sent.

* I integrated into MLton.Process instead of MLton.Child.  This took a
  minor amount of reorganization to avoid the dependency cycle you
  noticed. 

* Cygwin does not use mmap by default.  However, I added a runtime
  switch, "use-mmap", that lets one enable mmap-based memory
  allocation on the command line.  This has not been well tested.

* I did lots of reformatting and MLtonization.

There are a few remaining issues.

* I tested on all our platforms using a Linux -> <platform> cross
  compiler.  Everything worked fine, except for Cygwin.  With the 
  CreateProcess-based implementation, the test code dies when
  attempting to reap the process.  The error is
    unhandled exception: SysErr: No child processes [child]
  With the fork-based implementation, the test code hangs.  I'd be
  interested to see if you get the same.

* It seems wrong that on Cygwin MLton.Process.create expects
  Windows-style paths, whereas the rest of Cygwin expects Unix-style
  paths. 

* At some point, the {searchPath: bool} argument to Process.create got
  dropped.  I guess the idea is that we could have a separate function
  "val which: string -> string" that looks in the path for an
  executable?

* I think there may be a little room for improvement in making the
  remember function more type safe.  I'll think about it.

Hopefully it won't be too hard to cleanup whatever bugs I introduced.