Review and refactor row_purge_vc_matches_cluster()

As discussed earlier in MDEV-38140, there is a scope to review and refactor row_purge_vc_matches_cluster().

At least the functions row_purge_vc_matches_cluster(), row_upd_replace_vcol(), and trx_undo_read_v_cols() will have to be carefully reviewed. If I understood it correctly, row_purge_vc_matches_cluster() is not making any effort to compare to the latest version of the secondary index record. However, when row_purge_is_unsafe() is comparing to non-virtual secondary indexes, it does consider the latest version of the clustered index record unless it is delete-marked.

Possibly, the entire logic of row_purge_vc_matches_cluster() had better be integrated in the function row_purge_is_unsafe()