It seems that higher level keytab access routines like
krb5_kt_get_entry are built on the lower level sequential get
operations.  And these entries delete the cursor when they are done.

krb5_get_init_creds_keytab is an even higher level operation.

It doesn't seem unreasonable to me to simply document this
behavior--that higher level op.erations will reuse the same cursor and
thus interrupt a sequential get.