Fix a really interesting problem found by Volker's conversion of sessionsetup SPNEGO...

Fix a really interesting problem found by Volker's conversion of sessionsetup SPNEGO to asynchronous code.

Normally clistr_push_fn() can depend upon cli->outbuf being
initialized by negprot and sessionsetup packets, and cli->outbuf[smb_flgs2] being
correctly set with FLAGS2_UNICODE_STRINGS when cli_setup_packet() is called. When
all the sessionsetups are async, then cli_setup_packet() is never called, the async
code uses cli_setup_packet_buf() - which initializes the allocated async buffer,
not the cli->outbuf one. So the first time clistr_push_fn() is called is from
libsmb/clidfs.c:cli_dfs_get_referral(), just after the connection and tconX.
In this case cli->outbuf has never been initialized, and cli->outbuf[smb_flgs2] = 0
so the DFS query pushes ASCII on the wire, which is not what we want :-).

Remove the dependency on cli->outbuf[smb_flgs2] in clistr_push_fn(), and
fake up a SVAL(cli->outbuf, smb_flg2) value using cli_ucs2(cli) function
instead, which has been initialized. We only care about the FLAGS2_UNICODE_STRINGS
bit anyway.

I don't think this is an issue for 3.5.0 as the sessionsetup is still
synchronous there, but Volker PLEASE CHECK !

Jeremy.