Consider Profile-Guided Optimiazation tuning

Hi!

Recently I applied Profile-Guided Optimization on RocksDB ang got interesting results on the `db_bench` (a benchmarking tool from RocksDB):

Release:
```
DB path: [/tmp/rocksdbtest-1000/dbbench]
readrandom : 8.306 micros/op 120392 ops/sec 180.002 seconds 21670999 operations; 8.4 MB/s (13694547 of 21670999 found)
```

Release + PGO:
```
DB path: [/tmp/rocksdbtest-1000/dbbench]
readrandom : 7.415 micros/op 134862 ops/sec 180.006 seconds 24275999 operations; 9.4 MB/s (15344319 of 24275999 found)
```

I think the same optimization technique can be useful for LevelDB as well.