do not decode result_string returned by set_password

[zarganum]

Field testing of set_password() with latest MIT and real AD introduced a potential decryption error of result_string which comes from server in encrypted format.
The decryption works fine with k5test.
The root cause of this problem remains unclear (may be a native library, cipher negotiation issue, etc).
Until it is resolved, the Python code at least should not crash.
Thus I propose to not decode the string and leave as is.

UPD:
The result_string may be a binary encoded structure in proprietary format returned as result_string. It can be decoded and i18n'ized by client, like kpasswd(1) does.
K5test result_string is always a UTF-8 string which is also possible.
Current implementation leaves decoding to the client so result_string is returned as a raw byte sequence.

[zarganum]

this PR is legit because result_string from a real AD... contains a binary struct ad_policy_info that has to be decoded separately. And this is conditional, result_string still may be just a UTF-8.
Oh well, you'd never get bored with K5 and AD.

[jborean93]

Thanks for picking up this issue and fixing it!