D860568 fixed a double free, but introduced a leak. If no
timezone is specified in the input string, timelib_fill_holes
would clone the one from m_time. Since we unconditionally cloned
*that* into m_tz, the clone from timelib_fill_holes was leaked.
The contract seems to be that m_time doesn't own it's tz_info, but
that it should be owned by a smart pointer somewhere. So we can tell
timelib_fill_holes *not* to clone the tz_info from its second parameter,
and at the same time optimize for the usual case where that timezone
is already the current timezone.
mwilliams
bc5778169e