Migrating mprans to xtensor

[JohanMabille]

This issue is meant to track the migration of the mprans module to xtensor. This will be splitted in many steps to ease the merge of PRs and avoid having too many conflicts.

• Replace C-style arrays with pyarray in the C++ api exposed to Python
• Replace all the arguments with a dictionnary of arguments
• Move implementations to cpp files
• Propagate pyarray to the implementation
• Simplify the implementation based on xtensor API.

[cekees]

Thanks or this @JohanMabille . I know we discussed the dictionary of arguments approach. Do you have a prototype of this yet, say for calculateResidual of RANS2P?

[JohanMabille]

Not yet, binding a map of pyarrays to a dictionnary of numpy arrays does not prevent the copy of numpy arrays when called from the Python side. I know how to fix that but it won't be immediate, thus my question in #1068

[cekees]

@tridelat, can you take a shot at helping @JohanMabille add the xtensor dependencies to the stack as proper hashdist recipes and remove get_xtensor.sh? I think we should do this before the release. Here's an example of a recipe that copies header files to the artifact direction in the install target: https://github.com/erdc/stack/blob/master/pkgs/daetk/daetk.yaml

/cc @zhang-alvin

[tridelat]

@cekees @JohanMabille see erdc/stack#43