the previous commit 235d61890512894e27f4f81e38a325eee3c67b30, fixed just
exactly the problem reported in tdf#116981.

This commit fixes similar issues that may exist elsewhere.

To recap, this started as a regression from
        commit  433fc221 (patch)
        sal_uIntPtr->sal_Int32 in MultiSelection

Previously, MultiSelection stored it's values internally as sal_uIntPtr,
but returned them as long in FirstSelected(), NextSelected(),
and SFX_ENDOFSELECTION was defined to be ULONG_MAX.
On 64-bit Linux, sal_uIntPtr is typedefed to sal_uInt64, and ULONG_MAX
is 2^64, which means that previously, the SFX_ENDOFSELECTION value was
being converted from 2^64 to -2^63 when it was returned, which was why
these loop worked.

So convert SFX_ENDOFSELECTION to SAL_MIN_INT32, so we get a large
negative value which can never be a valid index, and which works more
like it did before the regression.
Also fix as many loops as I can find, to check against
SFX_ENDOFSELECTION explicitly.

Change-Id: I947d43dbe23a08105be3d849e33d7e774a8a19fa
Reviewed-on: https://gerrit.libreoffice.org/52934Tested-by: Jenkins <ci@libreoffice.org>
Reviewed-by: Noel Grandin <noel.grandin@collabora.co.uk>