zcore: build user EntryIDs from the whole address

[ximalin]

zcore mints ESSDN EntryIDs that gromox itself cannot resolve back. Measured with cvt_username_to_essdn() followed by cvt_essdn_to_username():

input	ESSDN	resolves back to
sender@example.org	.../cn=2200000011000000-sender	sender@example.org
sender -- what the caller passes	.../cn=2200000011000000-public.folder.root	ecUnknownUser

Mechanism

common_util_username_to_entryid() cuts the domain off first, but cvt_username_to_essdn() splits the address itself, and a name with no @ selects the public-store branch da7de4c added. The numeric half of the cn still names the right user, so nothing fails where the EntryID is written; cvt_essdn_to_username() is the reader that checks the textual half, and it answers ecUnknownUser.

This is the username_to_entryid callback zcore installs for iCalendar import (cu_ical_to_message, cu_ical_to_message2), so it covers the organizer (PR_SENT_REPRESENTING_ENTRYID, PR_SENDER_ENTRYID) and every attendee of an appointment created over CalDAV or from the web client.

Where it comes from

common_util_username_to_entryid() composed the ESSDN itself with a single snprintf() until 14aacee replaced that with a call to the shared helper and kept the caller's split in place. da7de4c had landed twelve days earlier, so the public-store branch was already there to absorb the bare name; before it the helper answered ecInvalidParam, and the same mistake would have failed loudly. The two other call sites in this file pass the helper a whole address.

Change

Hand username to the helper instead of tmp_name. This also drops the early return FALSE for a name with no @ in it: the helper handles that shape deliberately, and mysql_adaptor_get_user_ids() has already accepted the name.

Test

tests/utiltest gains the round trip for a full address and the public-store branch for a bare name, so the requirement on the caller is written down somewhere. make check passes 3/3.