Previously, if a controller wrote a flow with action NXAST_LEARN2, then
OVS would internally store an ofpact_learn structure with the raw type
set to NXAST_LEARN. When re-encoding, if the learn action happened to
have a limit or dst_ofs specified (which can only be specified for
NXAST_LEARN2), then it would re-encode using NXAST_LEARN2. However, if
these fields were both zero then OVS relies on the ofpact 'raw' type to
re-encode the action, so would end up encoding it as NXAST_LEARN in
subsequent serialization.

Fix this issue by storing the raw type when decoding learn actions.

VMWare-BZ: #1897275
Fixes: 4c71600d

 ("ofp-actions: Add limit to learn action.")
Reported-by: Harold Lim <haroldl@vmware.com>
Signed-off-by: Joe Stringer <joe@ovn.org>
Acked-by: Yi-Hung Wei <yihung.wei@gmail.com>