- The eliashberg_product_fft now uses indices instead of interpolation.
But this comes with the cost of giving a Gamma with the correct wmesh.
The sum over omega for the constant part was substituted by already
known Fourier transformation.
Also fixed 10.

- We now use semi-randomized inputs for the eigenvalue solver to get rid
of the tail fit problems.

- Restructure the eliashberg module to be more flexible
- Add power method
- Test eliashberg_product and the eigenvalue solver individually in
seperate tests.
- Add documentation

The constant part was included in matrix product like summation.

Also:
- use this macro in cmake files
- add cmakefile for eliashberg tests
- add matrix_rpa.py as a module
- minor eliashberg test changes
- adjust to new folder hierarchy

- If the plot is done in eliashberg/product_summation_vs_fft is now
regulated by a parameter.
- The solve_eliashberg method does now take initial deltas as input to
start the iterative solvers. It only uses the one produced by
semi_randomized_initial_delta if none is given to it.

- Add a first implementation for bandstructure like plotting WIP
- Add a matplotlib style sheet to use with the notebooks
- Add a user guide that uses the linearized Eliashberg equation WIP

If the dynamic part of the pairing vertex is zero the solving of the
eliashberg equation will only be done with the constant part reducing
the computational cost.

test:
- the eigenvectors can differ in sign depending on when the
eigenvaluesolver stops. Therefore check for sign flip.
- missing parenthesis

eliashberg.py:
- stopped using variable `eliashberg_product` or multiple things

- minor change in plotting_tools.py so it does not access the [0,0] of
the Gf, object

Lose the unnecessary model class and make `SquareLattice` a function
called `create_square_lattice`.

- norb is now the number of orbitals EXCLUDING spin to make the function
less error prone.

- high symmetry points are no longer part of the bsplot implementation
and the user gives now the k-points himself.
- axes tuning added
- dos implementaion added

This reverts commit 88989a3.

The two-particle Green's functions do not require to have a omega and
momentum mesh but can have an arbitrary one.

Instead of using loops over the indicies we are reordering indices to
use matrix products. This speeds up the construction significantly.
Also the singlet and triplet function now call the same creation
function but with different factors for charge and spin.

Adds a function that checks if two eigenvectors are equal if multiplied
by a scalar.

Make test more generic by variable hopping rates.
Also handle exceptions in model parameter lookup and slightly adjust
tolerance in testing.

The function `eliashberg_g_delta_g_product` and the dynamic part
in `eliashberg_product_fft` were refactored due to issue #725 of
TRIQS.
Also their outerloop was parallelized with openmp.

Because the `make_gf_from_fourier` function of TRIQS is not
parallelized for mesh products helper functions were introduced.
This is a overlap with some functions in `gf.cpp` and will be
changed later.

Added a template that uses the omp parallelized fourier
transformation for Gfs with a general target in fourier_gf.hpp
This template was then used is gf.cpp and chi_imtime.cpp
where duplicate definitions of a parallelized fourier
transformation existed.

The naming a placement of fourier_gf.hpp, or further
generalizations are up for debate.

We no longer compare signed to unsigned integer.

The parallelized template fourier function can now not only be used by
`const_view` types, but by any Gf type.
Also these fourier functions are now used to prepare Gamma in
`eliashberg.cpp` and the numerical tolerance was eased in a test.

Functionality of `eliashberg_product_fft` was pulled out and
put into a seperate function.
The function mpi_view is now used again to create arrays
for multi threaded enviroments.

One can now use the functions `enforce_symmetry` and `check_symmetry` one-
particle fermionic Green's function with a MeshProduct with a MeshImFreq on
first and a MeshBrillouinZone on second position to symmetrize or check
the momentum, frequency or orbital space.

The negative/positive half of the frequency were not mirrored before, resulting
in a incorrect enforcing and detecting of frequency symmetry.

In the momentum symmetry check numerical zero elements could trigger a +1 for
even symmetry, even through they don't matter in a symmetry check.

This is now fixed.

The sum over the fermionic Matsubara frequency i\nu for
\chi(i\omega, i\nu, q) can directly be calculated with the
function `imtime_bubble_chi0_wk`. This saves memory, because
the \chi(i\omega, i\nu, q) object does not need to be allocated
for the generally rather big number of i\nu points used for the
tail correction.
Still, \chi(i\omega, i\nu, q) needs to be allocated for the
number of i\nu used for \Gamma(i\omega, i\nu, i\nu) where memory
can be a restriction.

This showed, that there was a missing factor of 1/2
in the constant part of the particle-particle gamma.
Luckily, this is very insignificant, as I have checked
for previous calculations. Anyways, I have adjusted the
data for the previous implementation checks.

That FFT and SUM yield the same result is covered by
a different test.
Speed up from 15 sec to 2 sec.

The test 'symmetrize_gf' does already test the
functionality of the 'enforce_symmetry' function.
Therefore we now only test if 'solve_eliashberg' actually
uses this function and applies it correctly for two
different cases.
Speed up from 8 sec to 2 sec.

Speed up from 5 sec to 2 sec.

Speed up from 5 sec to 1 sec.

Speed up from 5 sec to 1 sec.