Fix compiler errors and warnings

[kimwalisch]

This pull request fixes the compiler errors and warnings further down that I encountered in my Pseudosquares-Prime-Sieve. I have tested these fixes on the montgomery_pow_kary branch of my Pseudosquares-Prime-Sieve project and all unit tests pass successfully: https://github.com/kimwalisch/Pseudosquares-Prime-Sieve/actions/runs/15957320257

I tested using Clang on macOS, using GCC on Windows with MSYS2/MinGW-w64 and using GCC and Clang on Linux. My code compiles without any compilers warnings now using -Wall -Wextra -pedantic.

Fixed errors

The errors below have been fixed by using if constexpr() from C++17. I have also updated the CMakeLists.txt and set cxx_std_17 on the hurchalla_modular_arithmetic interface library so that the compiler uses a C++ standard >= 2017 (when your library is included by some other CMake project). My error fixes require C++17 or later because I currently don't know how to fix these errors without C++17.

In file included from /Users/kim/GitHub/Pseudosquares-Prime-Sieve/src/pseudosquares_prime_sieve.cpp:19:
In file included from /Users/kim/GitHub/Pseudosquares-Prime-Sieve/src/modpow.hpp:23:
/Users/kim/GitHub/Pseudosquares-Prime-Sieve/lib/modular_arithmetic/montgomery_arithmetic/include/hurchalla/montgomery_arithmetic/detail/experimental/montgomery_pow_kary.h:119:23: error: static assertion failed due to requirement 'TABLESIZE % 64 == 0': 
  119 |         static_assert(TABLESIZE % 64 == 0, "");
      |                       ^~~~~~~~~~~~~~~~~~~

/Users/kim/GitHub/Pseudosquares-Prime-Sieve/lib/modular_arithmetic/montgomery_arithmetic/include/hurchalla/montgomery_arithmetic/detail/experimental/montgomery_pow_kary.h:100:9: warning: array index 16 is past the end of the array (that has type 'V[16]' (aka 'hurchalla::detail::MontyQRValueTypes<unsigned long long>::V[16]')) [-Warray-bounds]
  100 |         table[16] = mf.square(table[8]);

Fixed warnings

The warnings below have been fixed by using int instead of size_t for the P variable. I could also have casted P to an int but since this is basically required for every single usage of P it seemed better to simply change P's type to an int.

In file included from /Users/kim/GitHub/Pseudosquares-Prime-Sieve/src/modpow.hpp:23:
/Users/kim/GitHub/Pseudosquares-Prime-Sieve/lib/modular_arithmetic/montgomery_arithmetic/include/hurchalla/montgomery_arithmetic/detail/experimental/montgomery_pow_kary.h:288:36: warning: comparison of integers of different signs: 'int' and 'const size_t' (aka 'const unsigned long') [-Wsign-compare]
  288 |     HPBC_ASSERT(0 < shift && shift < P);

/Users/kim/GitHub/Pseudosquares-Prime-Sieve/lib/modular_arithmetic/montgomery_arithmetic/include/hurchalla/montgomery_arithmetic/detail/experimental/montgomery_pow_kary.h:141:25: warning: comparison of integers of different signs: 'int' and 'const size_t' (aka 'const unsigned long') [-Wsign-compare]
  141 |     HPBC_ASSERT(numbits > P);

[kimwalisch]

For your information:

I have now further optimized the math formulas in my Pseudosquares-Prime-Sieve project and using your montgomery_pow_kary() implementation is now slower than the default montgomery implementation for 128-bit numbers on all my three CPUs: Apple M2, AMD EPYC Zen2 and Intel Core Ultra 5 245K. Hence I will not use montgomery_pow_kary() in my project for the time being.

I do however use your montgomery_two_pow() implementation now, as I found it consistently speeds up my program by a small amount.