Do not append colon if uri password is NULL

[jarrodb]

I agree with @brettnem's assessment of the problem with #593. As @liviuchircu pointed out, this pull requests addresses the problem in do_action.c:1032 where a colon is blindly added if the user exists without consideration for the presence of an actual password... it attempts a memcpy on a NULL pointer no less (albeit with a length of 0).

I think if parse_uri(tmp, len, &uri)<0) finds an actual semi colon after the user and the password IS indeed blank, uri.password.s should be initialized to an empty string "" instead of NULL so that do_action.c:1032 check passes and the current "problematic" behavior will occur. But otherwise, if the password is NULL, the colon should not be appended.

This should satisfy the userinfo requirements as well as fix the issue presented.

Thoughts?

[liviuchircu]

Although current behavior barely passes RFC requirements (i.e. it's correct), a SIP server should try to be as permissive as possible when receiving, yet strictly adhere to recommended RFC behavior when sending. And appending unnecessary colons to URI usernames doesn't seem to be too recommended.

The fix looks great, since it still propagates existing empty-string passwords, if script writer really needs this. Many thanks, @jarrodb !