profiling

[Matthew Fluet]

> Now in the mlton executable, for the -p case, we still want  to  set  keepCps
> true.   We only need to set keepGenerated to true if information still has to
> be extracted from the C files, and I don't think that that needs  to  be  the
> case  if  Matthew  encodes what he needs in the static symbol names.  

I never look at any of the generated files, so we don't need keepGenerated
to be true.  Some of the labels that get spit out by mlprof will only make
sense after looking at the .s files, but if someone wants to read through
those files, they can easily compile with -keep g.

> Matthew, if there seems to be something wrong with this, I'm open to changes.
> Also if you don't want to write the thing that scans the resulting mlprof.out
> file, tell me what symbols you are including in your generated code and I can
> do that part.

The file format and all seems to be fine.  It should only take a couple of
hours (if not less) to fix up the newer version of mlprof.  I'll let Steve
integrate this profiling support, and if he can post a snapshot this
afternoon, I should be able to polish of mlprof in the early evening.