Add capability of saving GPU statistics and printing diff vs saved stats

[abouteiller]

1. new API parsec_devices_save_statistics fills up an array of statistics for later use
2. new API parsec_devices_print_statistics prints the diff between start_stats (as obtained in 1.) and the current internal stats
3. new API parsec_devices_free_statistics deallocate the array obtained from save_statistics.

This is to support customizable reporting of stats, see ICLDisco/dplasma#89

[devreal]

Is this a public API? Are we removing an old public API, breaking existing codes? Can both coexist?

[abouteiller]

The new 3 APIs in the .h are public, the old 2 APIs were private (and still exist, called from parsec_finalize)