The change N7 in MDEV-15340 (see the commit message) introduced
a regression in how CAST(AS TIME), HOUR(), TIME_TO_SEC() treat datetimes
'0000-00-DD mm:hh:ss' (i.e. with zero YYYYMM part and a non-zero day).
These functions historically do not mix days to hours on datetime-to-time
conversion. Implementations of the underlying methods used get_arg0_time()
to fetch MYSQL_TIME. After MDEV-15340, get_arg0_time() went through the
Time() constructor, which always adds '0000-00-DD' to hours automatically
(as in all other places in the code we do mix days to hours).

Changes:
1. Extending Time() to make it possible to choose a desired way of treating
   '0000-00-DD' (ignore or mix to hours) on datetime-to-time conversion.
   Adding a helper class Time::Options for this, which now describes two aspects
   of Time() creation:
   1. Flags for get_date()
   2. Days/hours mixing behavior.

2. Removing Item_func::get_arg0_time(). Using Time() directly
   in all affected classes. Forcing Time() to ignore (rather than mix)
   '0000-00-DD' in these affected classes by passing a suitable Options value.

3. Adding Time::to_seconds(), to reuse the code between
   Item_func_time_to_sec::decimal_op() and Item_func_time_to_sec::int_op().

4. Item_func::get_arg0_date() now returns only a datetime value,
   with automatic time-to-datetime conversion if needed. An assert was
   added to catch attempts to pass TIME_TIME_ONLY to get_arg0_date().
   All callers were checked not to pass TIME_TIME_ONLY, this revealed
   a bug MDEV-15363.

5. Changing Item_func_last_day::get_date() to remove the TIME_TIME_ONLY flag
   before calling get_arg0_date(). This fixes MDEV-15363.