Ceph

If cephfs client is old enough (for example Linux kernel 5.4 cephfs client) and uses struct ceph_mds_request_head_legacy,
then MDS server won't handle ext_num_retry, ext_num_fwd, owner_uid, owner_gid fields properly. They are left uninitialized
which leads to serious issues.

- In case of owner_uid, owner_gid it leads to the MDS crash on assertion [ ceph_assert(_inode->gid != (unsigned)-1) ].
- For ext_num_retry, ext_num_fwd it's hard to notice any issue but from the code I can see that issue is present. (I hope that Xiubo Li will check me there.)

Issue was originally discovered by Venky Shankar https://github.com/ceph/ceph/pull/53139#issuecomment-1771388202 during testing of the Quincy release.

After some investigation I was able to find a bug in the code, then reproduce it on my local environment and prepare the PR for the main branch (https://github.com/ceph/ceph/pull/54149).

But we will need to (partially) backport this to older cephfs releases.