Actually test GPU in test_boundary_handling.test_simple

See merge request pycodegen/lbmpy!25

Add test to compare results from OpenCL, CPU, CUDA in a full LBM simulation

See merge request !26

Fix pip installation

See merge request !27

For some reason their result does not agree with the calculated reference velocity.
Did not check for the analytical solution.

Was only allocating buffers for GPU but then doing boundary handling on
CPU

- corrected sorting 0.2.10 > 0.2.2

Enable iterative initialization for target='opencl'

See merge request !24

CI: Replace minimal-ubuntu job with ubuntu

Closes pystencils#19

See merge request pycodegen/lbmpy!22

- works on newer VS compiler locally
- comment this test out until CI system is properly updated

- test_serial_scenarios already tested if waLBerla is available by catching ImportErrors
- importorskip from pytest throws another exception, which is not part of the public API
-> catching all exceptions - which is not really elegant but works

- also for default MRT methods its better to use symbolic RR's instead of inserting numeric values for them
- but then the simplification strategy developed for SRT/TRT has to be disabled
- this commit introduces two new options:
     - simplification='auto' or a custom SimplificationStrategy
     - 'keep_rrs_symbolic'=True, by default now RRs are left symbolic

Execute LBM scenario with OpenCL

See merge request !13

Add custom timeloop_creation_function for LatticeBoltzmannStep

See merge request !20

Skip some tests if waLBerla is not installed

See merge request pycodegen/lbmpy!18

update of the kernel equations for the phase field model

See merge request !19

minor documentation fix for Shan-Chen

See merge request !21

are much faster in certain setups