Discard results if compilation happens during microbenchmark timing

```
http://stackoverflow.com/questions/15834059/caliper-micro-and-macro-benchmarks/1
5905931 mentions it at the bottom of the question. We've discussed it before
and it seems reasonable, but substantially different from Caliper 0.5 behavior.
```

Original issue reported on code.google.com by `gk5...@gmail.com` on 9 Apr 2013 at 3:26