Address various compiler warnings

At least the missing `return` statements can lead to undefined behavior, and in particular to crashes with clang.

See merge request !31

Merge branch 'feature/fix-missing-return-statements' into 'master'

Add missing return statements

This avoids undefined behaviour. However, I have no idea,
if `0` is a proper value to return.

Closes #14

See merge request !29

See merge request !30

Add missing return statements

This avoids undefined behaviour. However, I have no idea,
if `0` is a proper value to return.

Closes #14

See merge request !29

Fix/remove include directories

See merge request !26

It caused -Wmissing-include-dirs warnings in dune-uggrid

Update test environments

- No longer test with clang 3.5.
- Add clang 3.8 from Debian 8 + backports
- Add gcc 5.4 + clang 3.8 from Ubuntu 16.04 LTS.
- Use `duneci-standard-test`

See merge request !28

- No longer test with clang 3.5.
- Add clang 3.8 from Debian 8 + backports
- Add gcc 5.4 + clang 3.8 from Ubuntu 16.04 LTS.
- Use `duneci-standard-test`

[cmake] Remove checks for X11 (xlib)



See merge request !27

Remove more unused parts

This merge request removes a few more commands and some parts of `np/*`.

See merge request !24

DUNE uses the `newformat` command, but calls `CreateFormatCmd` (part of
`formats.h`) directly.

DDD's Fortran interface was already removed in
b49fe96e. This file was just missed.

It is not used anywhere.

Ensure a side vector's `object` and `VECTORSIDE` are consistent

A side vector belongs to one or two elements. One of them is stored as a
representative in the vector's `object` member and the side of the
`object` it belongs to is stored in `VECTORSIDE` as part of the control
word.

When restoring consistency in `ElementObjMkCons` this reverse link from
the side vector to the element is restored, but the `VECTORSIDE` was
not. This can lead to an inconsistent view for side vectors belonging to
two elements:

On the master, let the vector `v`'s representative element be `A` and
the side of the element be `a`. Let `B` (`b`) be the other
representative and assume `a ≠ b`. As the vector's control word is
global data (`EL_GDATA`), the non-master side vectors will also have
`VECTORSIDE` set to `a`.

If now `ElementObjMkCons` is called first for `A` and then for `B`, then
`v.object` will first be set to `A`, but then `B`. However `VECTORSIDE`
is still `a`!

This change makes sure that `VECTORSIDE` is also updated alongside the
`object` pointer. Note that it is (still) not guaranteed that the same
representative is chosen.

Closes #12

See merge request !23

Up to now, the version of the dune-uggrid module was 3.13.0, because that
was the version of UG when we forked it.  However, since dune-uggrid and
dune-grid are not separated by official interfaces, the two modules
practically have to be developed in lock-step.  Therefore starting with
this patch the dune-uggrid module will follow the release cycle and
the version numbering of the Dune core modules.

A side vector belongs to one or two elements. One of them is stored as a
representative in the vector's `object` member and the side of the
`object` it belongs to is stored in `VECTORSIDE` as part of the control
word.

When restoring consistency in `ElementObjMkCons` this reverse link from
the side vector to the element is restored, but the `VECTORSIDE` was
not. This can lead to an inconsistent view for side vectors belonging to
two elements:

On the master, let the vector `v`'s representative element be `A` and
the side of the element be `a`. Let `B` (`b`) be the other
representative and assume `a ≠ b`. As the vector's control word is
global data (`EL_GDATA`), the non-master side vectors will also have
`VECTORSIDE` set to `a`.

If now `ElementObjMkCons` is called first for `A` and then for `B`, then
`v.object` will first be set to `A`, but then `B`. However `VECTORSIDE`
is still `a`!

This change makes sure that `VECTORSIDE` is also updated alongside the
`object` pointer. Note that it is (still) not guaranteed that the same
representative is chosen.

Closes #12

[CMake] Don't add include paths that are no longer provided

This fixes missing-include-dir warnings in (some) dune-uggrid
and (all) dune-grid

See merge request !22

This fixes missing-include-dir warnings in (some) dune-uggrid
and (all) dune-grid

[cmake] fix install directory for DuneUggridMacros.cmake

The install directory has to be ${DUNE_INSTALL_MODULEDIR}. Otherwise,
the file will not be found by dune_process_dependency_macros from
DuneMacros.cmake.

See merge request !21

The install directory has to be ${DUNE_INSTALL_MODULEDIR}. Otherwise,
the file will not be found by dune_process_dependency_macros from
DuneMacros.cmake.

Test the FIFO implementation

This merge requests adds a simple unit test for UG's FIFO implementation.

See merge request !18

Remove several unused commands and some other minor cleaning up

This merge requests removes several commands that are no longer of interest, mostly related to the online documentation system and time measurement.

In addition some smaller cleanups:
- The unused `low/ugmemory.h` header was removed.
- Checks for standard C functions have been removed from `CMakeLists.txt`
- The `bin/ugdemo` script was removed. It need demo applications that are not included in dune-uggrid.

See merge request !19

This commit removes checks for functions and headers mostly required to
exist in a standard C (or C++) environment such as the `stdlib.h` header
or various string-related functions.

Some of the checks were looking for functions that are non-standard, but
the function is not (or no longer) used in UG, for example the test for
the (deprecated) `bzero()` function. These have also been removed.

These are part of the C standard.

Neither `fork()` nor `vfork()` is used anywhere in UG.