net/mlx5: support single flow dump

net/mlx5: support single flow dump

Modify API mlx5_flow_dev_dump to support the feature.
Modify mlx5_socket since one extra arg flow_ptr is added.

The data structure sent to DPDK application from the utility triggering
the flow dumps should be packed and endianness must be specified.
The native host endianness can be used, all exchange happens within
the same host (we use sendmsg aux data and share the file handle,
remote approach is not applicable, no inter-host communication happens).

The message structure to dump one/all flow(s):
struct mlx5_flow_dump_req {
	uint32_t port_id;
	uint64_t flow_ptr;
} __rte_packed;

If flow_ptr is 0, all flows for the specified port will be dumped.

Signed-off-by: Haifei Luo <haifeil@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>