...when adding or deleting a partition or setting its flag
rationale: telling kernel to reread partition table is done later in
partition_table::write() whereas partition_table::gpt::write() job is
just to update partition table on disk.

On some machines, calls to ped_disk_new() in libparted take of
the order of seconds, so doing this for each flag and partition
in turn makes partition_table::gpt::read_one take an inordinate
amount of time (mga#15621). Instead, collect the flags during the
call to c::get_disk_partitions.

just bind the libparted function (needed for next commit)

needed for next commit

ESP are the only partitions we're detecting through their pt_type
(modulo 0x12 which was Compaq Diagnostic & unlikely to be found on
GPT...)

However GPT has no equivalent to pt_type
But we can test for ESP flag

done separatly in order to have a better history for simplify then
introducing new API