include: share covered-switch pragmas [#3595]

[user01010111]

Summary

This moves the repeated GCC and Clang warning suppression around defensive
default: clauses into nut-pragmas-covered-switch-default.h and its paired
-end.h header. The headers are intentionally repeat-includable and are
distributed as private build inputs.

The change replaces 41 identical regions across 21 C and C++ translation
units. It does not alter any switch, case, or default body.

Related to #3595. This is one exact warning-family slice and does not close
the broader issue.

Scope

This deliberately leaves the following distinct cases local:

• include/nutipc.hpp
• the mixed -Wswitch-enum block in drivers/apcmicrolink.c
• eight GCC-only regions in drivers/tripplite_usb.c,
  drivers/powerp-bin.c, drivers/adelsystem_cbi.c, and
  drivers/generic_modbus.c

No other warning-suppression family is generalised in this PR.

Validation

• GCC 16.2.1, hard warning profile with -Werror: full clean build passed.
• Clang 22.1.8, medium warning profile with -Werror: full clean build
  passed.
• All 21 changed consumer files compiled in both builds, including the
  available Modbus, Avahi, neon, libusb, nutconf, C, and C++ paths.
• CppUnit passed with GCC and Clang: 34 tests each.
• Aggregate test suites passed with GCC and Clang: 9 of 9 each.
• NIT passed: 26 passed, 0 failed, 0 skipped.
• GCC and Clang C and C++ probes confirmed repeated inclusion and restoration
  of the warning state after each closing include.
• make stylecheck, make check-source-nonascii, and whitespace checks
  passed.
• A bounded distcheck-light run with CHECK_FILES_QUICK_TARGETS= and
  SPELLCHECK_ERROR_FATAL=no passed, including packaging, install,
  uninstall, and distclean checks. The candidate-specific quick checks were
  run separately as listed above.

The Clang hard warning profile stops on 14 warnings treated as errors: ten
existing conversion warnings in common/common.c and four existing global
constructor warnings in common/nutwriter.cpp. Compiling pristine source at
the same revision reproduced the same diagnostics.

An ordinary distcheck-light run is also stopped by existing Aspell findings
in unchanged NEWS.adoc and docs/man/apcmicrolink.txt. Those files match
the base revision byte-for-byte.

Local validation did not exercise Powerman, IPMI, GPIO, Modbus RTU USB,
macOS, or Windows paths. Upstream CI remains responsible for broader platform
coverage.

AI assistance

OpenAI Codex gpt-5.6-sol was used for repository analysis, implementation,
review, drafting, and validation with high reasoning. The human contributor
remains responsible for the contribution.

[github-actions]

A ZIP file with standard source tarball and another tarball with pre-built docs for commit 2978188 is temporarily available: NUT-tarballs-PR-3599.zip.

[AppVeyorBot]

✅ Build nut 2.8.5.5163-master completed (commit 756e15e844 by @)

• artifacts
• A 7-zip archive with built NUT for Windows debug binaries is temporarily available: NUT-for-Windows-x86_64-SNAPSHOT-2.8.5.5163-master.7z

[AppVeyorBot]

✅ Build nut 2.8.5.5166-master completed (commit 96d7de5c85 by @jimklimov)

• artifacts
• A 7-zip archive with built NUT for Windows debug binaries is temporarily available: NUT-for-Windows-x86_64-SNAPSHOT-2.8.5.5166-master.7z

[AppVeyorBot]

✅ Build nut 2.8.5.5167-master completed (commit f816b2d532 by @jimklimov)

• artifacts
• A 7-zip archive with built NUT for Windows debug binaries is temporarily available: NUT-for-Windows-x86_64-SNAPSHOT-2.8.5.5167-master.7z

[biergaizi]

Strongly approve this PR.

These pragmas were added during a project refactor campaign called "fightwarn" with the goal of suppressing all compiler warnings. Unfortunately, because it was rushed, the fixes themselves are often "brute-force" in nature. If the compiler says there's an integer overflow, the fix was to cast all possible integer values instead of fix the location of the overflow. If the compiler says there's a buffer overflow (even to a compiler false-positive), the fix was to add range-checked in all locations where the value is used, rather than fixing the logic.

For example, one commit was introduced to the driver I maintain to suppress an integer overflow warning:

crc16_recv = (uint16_t)((uint16_t)(ident_response_end[0]) << 8) | (uint16_t)(ident_response_end[1]);

I later reverted this change because it was clearly an overfix.

These GCC pragmas are my most disliked commits during the fix. If these changes were subject to review of the maintainer of the respective drivers, I would have rejected them in my driver immediately. Sharing them into a standard header is a choice that I would make from the beginning, if the I don't understand why the copy-and-paste approach was used in the first place.

[jimklimov]

I dislike the screenfulls of these pragmas sprinkled over the code too, and kick myself for not thinking about the include-file approach earlier :)

As for keeping the warnings quiet, they are mostly (hopefully) due to sort-of-overkill in range checks that may be pointless in specific-bitness builds, but may vary between platforms (OTOH, like checking int i <= INT16_MAX which depends on bit-width of a native int, maybe always true, maybe not).

I hope there was not too much brute-forcing per se, although it is possible that attention meandered and some corners got cut as hundreds of very similar cases were addressed by hand or mass replacement patterns...