Added separate CI and CD workflows

[tdulcet]

This backports many of the same improvements that I made to PRPLL-NTT in gwoltman/gpuowl#7 and gwoltman/gpuowl#8.

• Changed the existing CI/build workflow into a separate CD workflow
  • Changed to build the Linux executables on CentOS 7 and Rocky Linux 8 instead of Ubuntu
    • For CUDA 12.5-13.2, this lowered the required glibc version from 2.39 to 2.28.
    • For CUDA 8, this lowered the required glibc version from 2.23 to 2.17.
• Added new CI workflow
  • Eliminates the need to build all 48 combinations of OS and CUDA version on each commit and PR
  • Added tests with Clang on Linux
  • Changed to build the debug executables, as the CD workflow now builds the release executables
  • Added ShellCheck job
• Updated Makefile
  • Added new DEBUG flag
  • Enabled -ffast-math for consistency with the Windows Makefile
  • Removed obsolete and redundant optimization flags
• Updated Windows Makefile
  • Enabled compiler warnings
  • Enabled Link Time Optimization (LTO) for consistency with the Linux Makefile
  • Removed obsolete and redundant optimization flags
• Enabled GCC/Clang to check the printf style arguments to the logprintf() function calls
• Minor improvements to Bash scripts
  • Fixed ShellCheck errors
  • Added error checking to the build helper script
  • Removed unnecessary quoting

Also see: primesearch/mfakto#65

[ixfd64]

I don't see any issues.