fix compiler errors for modern linux+clang+python+cuda setup

[rawbby]

• fix deprecation errors with new cmake versions regarding how CUDA is loaded
  (by upgrading cmake and the associated CUDA configuration)
• fix several minor compiler errors and warnings appearing on clang (see changes)
• fix usage of unstable pybind dependency (that crashed my compilation)
  (by using git checkout after submodules are loaded)
• fix unsynchronised update of submodules (crashes cmake configuration if multiple configurations run in parallel)
  (by introducing a lock file)

These changes are tested on my local setup:

Ubuntu                          24.04.2 LTS
clang                           18.1.3
ninja                           1.11.1
cmake                           3.28.3
CUDA SDK                        12.6.3
CUDA C++ Core Compute Libraries 12.6.77
CUDA Runtime (cudart)           12.6.77
Python                          3.12.3

using -DCUDA=ON -DPYTHON=ON with build types Release and Debug.

BUT THERE WHERE SOME FAILING TESTS!

The following tests FAILED:
 54 - */expr.storage/* (Failed)
 99 - qcd.mul1 (Failed)
119 - scheduling_eval.spmvGPU (Failed)
124 - scheduling_eval.ttvGPU (Failed)
159 - scheduling.mergeby (Failed)
225 - */stride.windowing/* (Failed)
226 - */indexSetVectors.windowing/* (Failed)
227 - */indexSetMatrices.windowing/* (Failed)

As I did not change any logic in the code I assume these are not based on my changes but on my setup.

I understand that raising the cmake version is not appreciated by everyone, but modern cmake will not support version lower than 3.10 in the future.
As with 3.18 the CUDA support is improved a lot I upgraded not only to 3.10 but 3.18 which is reasonable in my opinion.