[siracozmen01]

Problem

A library's <PP>_NODISCARD used to take [[nodiscard]] in C only from __STDC_VERSION__ >= 202311L. GCC 14 reports its C23 mode as 202000L, so under GCC 14 the macro was empty and a program that dropped a status got no warning. Every library has now taken maul-locale's fixed block, but maul_add_header_tests checks the macro only in C++17, so a library that went back to the old test would pass every check.

Change

• maul_add_header_tests writes c23_nodiscard.c, which includes the library's base.h and asserts in C23 that <PP>_NODISCARD is not empty, beside the C++17 check. It builds in header_test_c23, so it is skipped where that target is (MSVC's cl).
• maul-conventions.md says how the macro reads C dialects and that the header tests check it in C23 and C++17.

Verification

The generated file, compiled with -std=c23 -Wall -Wextra -Wpedantic -Werror against each library's headers at its current main:

• all twelve libraries pass under gcc-14 and clang-20;
• maul-audio's main before its fix (#355) failed under gcc-14 ("MAUD_NODISCARD is empty in C23") and passed under clang-20, which reports 202311L.

Closes #121

[siracozmen01]

Review at 528d93b: ready. Read, not run.

c23_nodiscard.c is written in the same branch as the C++17 file, so it exists whenever base.h has a column-0 #define <PP>_NODISCARD [[nodiscard]]. That line stays in a library that goes back to the old >= 202311L test, so a regression is still checked. The generated file is the C++ one with "C23" in its message. It goes into maul_c23_sources, which header_test_c23 builds with C_STANDARD 23 and the family's -Wall -Wextra -Wpedantic -Werror. Like the per-header C23 files, it is left out only under MSVC's cl (CMAKE_C_COMPILER_ID STREQUAL "MSVC"). clang-cl still gets it through /clang:-std=c23. The check is independent of a C++ compiler, which matches the comment's "Without a C++ compiler the C++ checks are skipped".

The assertion can fail: an empty macro stringizes to "", size 1. That is the body's maul-audio pre-fix result under gcc-14.

The comment and the maul-conventions.md line describe the C rule as the libraries now write it. I did not run docs_check.py, since the local checkout is not at this head.