Update to cudawrappers 0.4.0 and fix build

This repository had cudawrappers 0.1.0 as dependency, but didn't build against it any more due to changes against a local (diverged header-only) copy of cudawrappers. To fix this, the cudawrappers submodule is updated to version 0.4.0, include directories and the Makefile are updated. Users are now required to build cudawrappers themselves, the README is updated accordingly.

Additionally, this MR adds CI. See also here.

external/cuda-wrappers

-Subproject commit 884fbc7c69617f0b4fbc6696435272a488f49716
+Subproject commit d7f133c7e10c238d4bcceb2219f658eec64e9d7f