Reduce the overhead of parallel I/O operations

[FrancescAlted]

During the parallel writes / reads in compression / decompression, there is currently a sequence of open - write/read - close operations for every block and thread. By opening the chunk only once per thread and storing the file handler on every thread, we could avoid the open and close operations and just issue read / write ops. This should provide better performance (albeit at the cost of code complexity indeed).