Else (on 64bits platform) the driver smashes our stack: SQL_ULEN is 64 bits
Widen result type of getQueryTimeOut, getMaxFieldSize and getMaxRows so that it will always fit

- as long as we still have the slow build.pl run for top-level build

GTK themes expect the button and the editing area to be swapped in RTL.

When entering pass count 2 it is expected that the breakpoint is
passed two times before breaking not to break the second time.

Also fixed a crash due to dangling pointers of breakpoints.

...with SAL_LOG=+INFO due to std::ostringstream in log.cxx using global operator
new, which potentially calls rtl_allocateMemory.

It is puzzling that Executable_soffice.bin.mk contains the line
$(eval $(call gb_Executable_set_targettype_gui,$(sofficebin),YES))
yet, in the build log we see -SUBSYSTEM:CONSOLE among the linker flags.
I added -SUBSYSTEM:WINDOWS flag exlicitely, and it comes later so it
prevails. I guess MinGW will be still affected. It really should be
fixed correctly by a gbuild expert.

I added that as a convenience method while working on implementing
multi-value filtering.  But we need to stick with the plural version
GetQueryItems() from now on for clarity.

We are still not 100% compatible, e.g. Excel's "show blanks" setting
is not cleanly mapped (partly due to ODF's limitation).  But that's
a future TODO.

+ IsValidDate() checks only day and month regarding the year, not Gregorian
  cut-off date as now does IsValidAndGregorian().
+ Normalize() carries over invalid day and month values to next months and
  years.
* All methods that return or internally use a day count now internally
  normalize the date values, without modifying the actual Date instance. So,
  if the date is not valid you may get unexpected results.
  * Previously, a date with month>12 would had accessed the days-of-month
    array out of bounds on all such methods. So you would had gotten
    unexpected results anyway..
  * Affected methods are:
    GetDayOfYear()
    GetWeekOfYear()
    GetDaysInMonth()
    static DateToDays()