Fix CI on develop (#5750)

* Fix CI configuration broken by recent merges and tool updates

- gcc_flags.cmake: drop -Wexperimental-fmv-target, which GCC 16 accepts
  only on aarch64; amd64 rejects it, so every GCC job failed while
  checking the compiler.
- ci_get_cmake: add VERBATIM so the checksum pipeline is passed to the
  shell intact (the unescaped `$'` broke the generated Makefile and
  build.ninja, failing ci_cmake_flags and ci_module_cpp20); match the
  SHA-256 entry case-insensitively, as CMake 3.5.0 lists the archive as
  "Linux-x86_64"; and unpack with --strip-components, as that archive's
  top-level directory is spelled "Linux" too.
- ci_single_binaries: compile json.hpp's TU without IWYU's --error, as
  the comment above the gate already intends.
- tests: restore -Wno-deprecated-declarations for all non-MSVC compilers
  (#5737 kept it for GCC only), as several tests call deprecated
  functions on purpose; include thirdparty/fifo_map as SYSTEM.
- .clang-tidy: set misc-use-internal-linkage.AnalyzeTypes to false;
  clang-tidy 22.1 extended the check to classes and enums and flagged
  100 test helper types.

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

* Fix library warnings and a JSON_DIAGNOSTICS parent bug

- binary_reader: pass integers to sax->number_integer() through
  conditional_static_cast<number_integer_t>, making the existing
  narrowing for a narrow number_integer_t explicit (MSVC C4244 and GCC
  -Wconversion/-Warith-conversion with the int16_t test from #5694);
  mark two Infer DEAD_STORE false positives with @infer-ignore.
- to_json: set the parents of an array built from a C++20 range view
  after all elements are in place; a reallocating push_back moved the
  earlier elements and left their parent pointers stale, failing the
  JSON_DIAGNOSTICS invariant assertion.
- json.hpp: suppress MSVC C4127 for the new is_ordered_map check in
  diff(), like the three existing ones; spell out std::formatter::parse's
  return and iterator types for clang-tidy 22.1.
- number_parse: make the Eisel-Lemire digit counter unsigned
  (GCC -Wstrict-overflow).
- string_utils: take encode_utf8's callable by const reference
  (cppcoreguidelines-missing-std-forward) and drop a \u from its doc
  comment (-Wdocumentation-unknown-command).
- ordered_map: include <memory> for std::allocator (cpplint).

Ran make amalgamate.

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

* Fix tests failing in CI on develop

- unit-allocator: skip the #5640 test under MSVC STL iterator debugging,
  where containers allocate a debug proxy in noexcept move constructors
  and a failing allocation terminates; move a decrement out of an if
  condition (bugprone-inc-dec-in-conditions).
- unit-conversions: expect the "(/0)" path with JSON_DIAGNOSTICS;
  compare strict enums via get<>() rather than through the noexcept
  operator==(ScalarType, json), which bugprone-exception-escape flags.
- unit-alt-string: suppress -Wexit-time-destructors for the strict enum
  macro and misc-use-internal-linkage for its enum.
- unit-bjdata: call the static lookup functions through the type and
  pass unsigned char (-Wsign-conversion on amd64).
- Mark Infer false positives with @infer-ignore in unit-diagnostics,
  unit-pointer_access, unit-udt, and unit-conversions.
- Smaller clang-tidy 22.1 findings in unit-class_parser,
  unit-constructor2, unit-custom-base-class, unit-locale-cpp, and
  unit-noexcept.

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

---------

Signed-off-by: Niels Lohmann <mail@nlohmann.me>

tests/src/unit-bjdata.cpp

@@ -67,10 +67,9 @@ TEST_CASE("BJData")
 {
     SECTION("binary_reader BJData lookup tables")
     {
+        // both lookups are static member functions
         std::vector<std::uint8_t> const data;
-        auto ia = nlohmann::detail::input_adapter(data);
-        // NOLINTNEXTLINE(hicpp-move-const-arg,performance-move-const-arg)
-        nlohmann::detail::binary_reader<json, decltype(ia)> const br{std::move(ia), json::input_format_t::bjdata};
+        using reader_t = nlohmann::detail::binary_reader<json, decltype(nlohmann::detail::input_adapter(data))>;
 
         // the excluded optimized-type markers must match binary_writer's
         // is_bjdata_excluded_type_marker(), which encodes the same 8 markers
@@ -78,13 +77,13 @@ TEST_CASE("BJData")
                 {'[', '{', 'S', 'H', 'T', 'F', 'N', 'Z'
                 })
         {
-            CHECK(br.is_bjd_excluded_optimized_type(marker));
+            CHECK(reader_t::is_bjd_excluded_optimized_type(static_cast<unsigned char>(marker)));
         }
         for (const char marker :
                 {'U', 'i', 'u', 'I', 'm', 'l', 'M', 'L', 'd', 'D', 'C', 'B', 'x'
                 })
         {
-            CHECK(!br.is_bjd_excluded_optimized_type(marker));
+            CHECK(!reader_t::is_bjd_excluded_optimized_type(static_cast<unsigned char>(marker)));
         }
 
         // every dtype marker must round-trip to its ND-array type name
@@ -96,11 +95,11 @@ TEST_CASE("BJData")
         };
         for (const auto& type : types)
         {
-            const char* name = br.bjd_type_name(type.first);
+            const char* name = reader_t::bjd_type_name(static_cast<unsigned char>(type.first));
             REQUIRE(name != nullptr);
             CHECK(std::string(name) == type.second);
         }
-        CHECK(br.bjd_type_name('x') == nullptr);
+        CHECK(reader_t::bjd_type_name(static_cast<unsigned char>('x')) == nullptr);
     }
 
     SECTION("individual values")