There were code paths within CreateLine where the method did return a value,
and others where it didn't.  No caller of CreateLine actually considered
the return values at all.  Hence we may as well decide to not return anything
at all.  This should fix an issue reported by the undefined-behavior sanitizer.

Thanks to Carsten Gräser for pointing this out.

To get rid of tabs, mainly.  These two files were left out when moving
to git, because uncrustify wouldn't take them.  With a bit a manual
arm-twisting uncrustify now processes these files.

The (very) old DDD debugger with the (very) old motif gui cannot handle umlauts.

Compilers are much smarter than programmers, and know best when to use
CPU registers.  Leave the decision to them and do not clutter the code
with the 'register' keyword.

This used to be legal, but is not anymore.  In particular, clang
refuses to compile it.  This patch fixes the issue.

The explicit list of 8 bit numbers in the files tool0, tool1, etc
were stored in arrays of char. However, they contain values larger than 127,
which is illegal in C++11 -- the type must be unsigned char.
On the other hand, the char arrays are handed to X, which really does
expect signed char.
The patch fixes the problem by changing the types to unsigned char,
and explicitly casting the arguments of the X calls.

The Chaco load balancing code has been remove years ago, but the code that
parses argument strings for Chaco arguments was still there.  This patch
cleans up the code, removing all obsolete load balancer options.

For each SideVector, the VCOUNT field stores how many elements reference
the SideVector (2 or 1).  The VCOUNT field is apparently not touched
during load balancing at all.  Therefore, if it was set properly before
load balancing it is most likely screwed up afterwards, as elements have
gained/lost neighbor elements.  This leads to strange segfaults much
later.  Most notable was Dune FlySpray issue 810:

https://dune-project.org/flyspray/index.php?do=details&task_id=810

which was hacked around in the UGGrid Dune-UG interface class as early
as 2010, in dune-grid commit 5e870eb04580e3279277ffdad917edf7c4dbae18

This patch here now fixes the issue 'properly', within UG itself.
It doesn't really do anything else than the UGGrid hack: after load balancing
has been completed, rewrite all VCOUNT fields.  After some investigation
I did not see any better way to obtain proper VCOUNT fields.  At least
the code is in the best spot for this now.  Namely it is part of the
method ConstructConsistentGridLevel.

It has been set ever since 1998.  I don't think we'll ever go back
to the old style.

They were used by the Chaco mesh partitioner, but the Chaco code has
been removed from the UG git tree a long time ago for license reasons.
The lb2 and lb3 fields didn't have any purpose after that, but
apparently this fact went unnoticed.  Now they have been removed,
which should save 8 Bytes per element.

Those were only used in the fortran backend, which has been removed.

This should fix a build failure on ppc64el:

+---
| wop.c: In function 'UG::INT OrderFathersXSH(UG::D3::MULTIGRID*, UG::D3::ELEMENT**)':
| wop.c:16705:60: error: invalid conversion from 'UG::INT (*)(const void*, const void*) {aka long int (*)(const void*, const void*)}' to '__compar_fn_t {aka int (*)(const void*, const void*)}' [-fpermissive]
|   qsort((void *)OE_Map, OE_nBndElem, sizeof(MAP), CompareIDs);
+---

Reference: https://launchpadlibrarian.net/174208169/buildlog_ubuntu-utopic-ppc64el.ug_3.10.0-1_FAILEDTOBUILD.txt.gz