Add constexpr (not for geometricMean which uses std::sqrt) and noexcept.
Static assert the requirements on the input arguments, they have to be numbers.

[discretization] Rename DiscretizationMethods to DiscretizationMethod with lower-case fields

Closes #444

See merge request !806

[assembly][newton] reimplement partial reassembly

Closes #379

See merge request !796

Bernd Flemisch authored 7 years ago and Timo Koch committed 7 years ago

Transfer the partial reassembly functionality to the current
structure.

The `NewtonSolver` is responsible for the partial reassembly.
It owns a `unique_ptr` to an object of a new class
`PartialReassembler` which is invoked in the constructor as well as
in the methods `solve`, `newtonUpdate` and `assembleLinearSystem`.
The `PartialReassembler` object is created only if the parameter
`Newton.EnablePartialReassembly` is set to true.

After each Newton iteration, the `PartialReassembler` colors the
geometrical degrees of freedom depending on the shift of the
primary variables from the last time that the derivatives have been
calculated. This shift is updated by the function
`updateDistanceFromLastLinearization_` of the `NewtonSolver` and
passed to the function `computeColors` of the `PartialReassembler`
for the actual coloring. For green-colored entities, the entries in
the Jacobian are kept, while for the other entities, they are
recalculated.

Use an enum class `EntityColor` for the coloring, enable read
access to the color of an entity via the `PartialReassembler`.

Pass a pointer to the `PartialReassembler` when calling the routine
`assembleJacobianAndResidual` of the `Assembler`. Use SFINAE and
`enable_if` so that only assemblers accepting the additional
argument are invoked. The correspnding local assemblers have to be
adapted such that no calculations are performed for green elements.
Implement the functionality for Box, TPFA and MPFA; adapt the
`FVAssembler` as well as the `Box`/`CCLocalAssembler`
correspondingly.

The `PartialReassembler` throws if constructed with a discretization
method equal to None or Staggered. All `BoxLocalAssembler`s apart
from the implicit numeric one throw if called with a non-nullptr
to a `PartialReassembler`.

Allow for a default `nullptr` parameter to the assembly functions
that take a `PartialReassembler`. This requires a new class
`DefaultPartialReassembler` to be passed as a default template
parameter for these functions. This makes it possible to use the
functions `assembleJacobianAndResidual` of the global and local
assemblers as before without the additional argument.

Allow to set the thresholds and the weight for the calculation of
the effective reassembly threshold as parameters. Try to explain
the corresponding formula in a comment.

Use the functionality in the two-phase incompressible tests.

[staggered] export the discretization method from the FVGridGeometry

See merge request !799

[headercheck] fix several headers to pass make headercheck

See merge request !800

Increase consistency with the other FVGridGeometries.

[typetraits] Implement isValid function to generate class member introspection functors

See merge request !797

* isValid can be used to generate a functor that can check whether an operation done with a type is valid
* this patch also provides examples in form of a test and documentation

Feature/variable eps

See merge request !791

[newton] use new NewtonSolver in incompressible 2p fv test

See merge request !795

[newton] format output consistently

See merge request !793

[newton] fix for use with Dune 2.5

See merge request !792

[nonlinear][NewtonSolver] Fix compilation with g++5

Closes #445

See merge request !794

There seems to be a bug in g++5 which which prevents compilation when
defining the call to the implementation directly to Dune::Hybrid::forEach.

We therefore store this call in a lambda and pass it to the
for loop afterwards.

Use a uniform number of digits for the iteration count as well as
a consistent format for the floating point values.

See !779.

Feature/unify newtoncontrollers

See merge request !762

[test][freeflowMaxwellStephan] Fix test

See merge request !790

* this is the only test that failed after the changes in momentum upwind, thus the use of inertia terms is disabled
* use fixed times to write output
* do some clean-up

[navierstokes] Improve flux variables

See merge request !745

Somehow the upwinding and changes in navierstokes led to an additional Newton step

* rename old test for low Re testing

* introduce upwind weight for momentum flux
* use convenience function to calculate all momentum fluxes
* rename computeNormalMomentumFlux to computeFrontalMomentumFlux
* rename computeTangetialMomentumFlux to computeNormalMomentumFlux
* move pressure term from  local residual to fluxVars
* use separate function to treat outflow boundaries
* improve docu and clean-up

* explicitly check for outflow BC
* small clean-ups
* improve docu

[impes] Remove grid singleton from problem

See merge request !773

* NewtonMethod
* NewtonController