core: replaced utils::strcat_alloc with printf

[reinaual]

Fixes #4020

Description of changes:

• replaced manual string char* construct with std::printf
• introduced verbose parameter to touched tuning functions and interfaces
• removed utils/strcat_alloc.hpp with unit tests
• added missing handle_errors call in p3m_gpu_init

[jngrad]

Here and elsewhere: I do not immediately see the advantage of using a template parameter instead of a regular parameter. Could you please let me know why this was necessary?

With a template function, there is no guarantee the true and false instantiations will do the same calculation. It is, for example, perfectly legal for someone to inject this snippet of code at line 798:

template <>
double p3m_mc_time<false>(const int mesh[3], int cao, double r_cut_iL_min,
                          double r_cut_iL_max, double *_r_cut_iL,
                          double *_alpha_L, double *_accuracy) {
  return 0.;
}

This will cause the tuning to select random values, which means we should duplicate the mmm1d/p3m/dp3m python tests to make sure verbose=False gives the same results as verbose=True.

[reinaual]

The template was just personal preference. I dont care, your choice is probably to change it to a regular paramter, i can quickly do that if you want me to.

[jngrad]

I think it would be more maintainable with a regular parameter. This would also allow you to merge p3m_adaptive_tune_core() and p3m_adaptive_tune() back into one function.

[reinaual]

im working on it, this is an artifact of cython not supporting non-type templates