Feature/fix mixed precision dense matrix vector interface

In the current state if the dune-common dense LA I encountered several smaller bugs w.r.t. mixed precision computations:

  1. the value_type used to reset the vector during `DenseMatrix::mv` and `DenseMatrix::mtv` was the `value_type` of the matrix, but it has to be the `field_type` of the vector.
  1. the type of the scaling parameter alpha of `usmv`. `usmtv` and `ushtv` has the wrong type and must be the `field_type` of the result vector. For details see bdf74dca

Besides this there were 2 minor issues I found while fixing the mixed precision case.

  1. the range checks we no longer performed
  1. if the tests are reenabled, they do not compile, because Robert added test for `FieldMatrix` with `std::vector` or C-array, which do not work, as we expect additional methods on the classes.

I enabled the range checks and disabled the `std::vector` and C-array tests. We could work around this problem, by adding helper traits to determine the size and by this allowing us to special-case `std::vector` and C-array.

See merge request !10

void_t, real_t, field_t in dune/common/typetraits.hh

Contains:
 - void_t (+voider)

 - template <class T>
   using field_t = typename FieldTraits<T>::field_type;

 - template &lt;class T&gt;
   using real_t = typename FieldTraits<T>::real_type;

Requires C++11.

See merge request !20

Rename vec_access() and vec_size() methods

This commit renames the `vec_access()` and `vec_size()` methods to match
the ones in the `DenseVector` facade class, that is `operator[]()` and
`size()`.

Closes #3

See merge request !12

added void_t (incl. voider-hack for gcc), and template aliases real_t and field_t for convenient access of T::real_type, T::field_type

added void_t (incl. voider-hack for gcc), and template aliases real_t and field_t for convenient access of T::real_type, T::field_type

Start hello world program in new projects with editor hints

Add editor hints from
https://www.dune-project.org/doc/devel/codingstyle.html in hello world
program created by duneproject.

See merge request !18

Feature/remove mpi deprecation warnings

Suppress some warnings due to deprecated methods in MPI. 

There are some warnings left due to MPI_LB and MPI_UB but I am not sure how to fix them. See http://www.mpi-forum.org/docs/mpi-2.2/mpi22-report/node76.htm.

See flyspray/FS#1610

See merge request !14

We require MPI 2.1 anyway and nobody has reported problems.

Feature/conceptcheck

This introduces the function

    template<class C, class... T>
    static constexpr bool models();

Its result indicates if the concept C is modelled by the collection
T... of types. A concept is a list of expressions that a type must
support. Concepts can refine other concepts to inherit all requirements
from those base concepts. Besides the concept check function itself, this contains
some helper functions of the form `require*` that transform
a non-matching condition to a substitution failure. E.g.,

    requireConvertible<R>(t.f())

is not valid unless the result of `t.f()` is convertible to R.

Since `models()` is the user interface, it's contained in the
namespace Dune::. In contrast the various `require*()` functions
needed for concept definitions are in the namespace Dune::Concept::.
I strongly propose that concept definitions themselves also go into
namespaces Dune::(NonCoreModule::)Concept.

This merge request depends on the TypeList feature (!7).


See merge request !9

flyspray/FS#1147: Compute norm of vectors with NaN entries properly

This is an attempt to solve the issue outlined in flyspray/FS#1147. I've yet to fix `DenseMatrix::infinity_norm` the same way.

See merge request !15

Test complex<T> for floating point T only

The standard leaves (the effect of instantiating) complex<T> for more
general T unspecified.

While we're at it, kill duplicate tests and add long double to the list.

This fixes bug flyspray/FS#1726: Now also with clang 3.7 and libstdc++ 4.9.2 on Debian 8.2, the test fvectortest can be compiled again.

See merge request !11

We require MPI 2.1 anyway and nobody has reported problems.