improve Kerberos name parsing and canonicalization testing

Currently we are not testing Kerberos name parsing. Kerberos name parsing is error prone because Keberos principals are complex; see http://web.mit.edu/kerberos/krb5-1.5/krb5-1.5.4/doc/krb5-user/What-is-a-Kerberos-Principal_003f.html.

Bugs such as https://issues.apache.org/jira/browse/ZOOKEEPER-1195 would have been caught, had we better tests. Although we cannot test (yet) a full end-to-end KDC realm, we can at least test Kerberos principal syntax and semantics.