Think about write_csv again

Performance for this is pretty embarrassing:

``` R
df <- as.data.frame(matrix(runif(256*65536), nrow = 256))

system.time(readr::write_csv(df, "df.csv"))
```

e.g. http://stackoverflow.com/questions/10505605/speeding-up-the-performance-of-write-table/36484339#36484339
