708def7d made that code conditional on !$do_link
so that it worked right for the case where making dev-install calls ooinstall
-l.  However, that code now starts to fail also in the --disable-linkoo case
(where building dev-install calls ooinstall w/o -l).  I have no idea whether
that code serves any real purpose in any actual use case, so remove it
completely for now.

When two (or more) text frames was imported without a non-frame
paragraph in between, the first frame was anchored to the second one,
instead of a non-frame paragraph.

The fix is modelled after what the old RTF import already did in
SwRTFParser::Continue() in swparrtf.cxx:493 and
SwRTFParser::SetFlysInDoc() in rtffly.cxx:481.

Otherwise the package creating code (part of the SDK) won't include
them and/or the package installation code (on the OS itself) won't
unpack them. (They just silently skip the file.)

Would it perhaps have been better to always link with gb_STDLIBS
without having to mention it in each Library_*.mk and Executable_*.mk?
And then for those few exceptions come up with some "No STDLIBS,
thanks" thing?

This reverts commit e0bea132.

XDatabase range test case is now finished

TODO: remove the explicit test file dependency from the interface test
code
don't use one of the test files for another class

Add smoketest to tail_build and make it a subsequentcheck target.

to equalsIgnoreAsciiCaseAscii("...")