pcp

pmlogreduce - fix 2 problems

1. a memory leak issue identified and triaged by SGI (thanks to
   Arthur Kepner and Jason Lim for this)

2. if no qualifying metrics were found in an output sample interval
   (this is something of a rare case for most use patterns), the code
   previously generated a bogus "mark" record ... this has been fixed
   and no output record is written when this happens