Sometimes the optimizer provides inconsistent parameter sets for the wop
PBS depending on how many inputs & bits to extract we're going to need
for a certain precision. This results in r < 0 and until now the
corresponding error in concrete-cuda was very hard to understand. This
commit fixes this behavior.
Also, when introducing the support for k > 1 we forgot to update the
checks on the low lat PBS. This commit also fixes them.
- Update concrete-cuda with latest commits in concrete-core
- Add C++ tests based on google test
- Run the tests in the CI with Cuda 11.8 and Cuda 11.1
- Test for many PBS/KS parameters
- Repetitions & samples are set for each parameter set in the PBS test
