feat(testing): add IsElementOf / IsKeyOf / IsValueOf matchers

[helly25]

Summary

gmock's Contains(value) puts the container on the left and the element on the right; the opposite orientation — is this value a member of that container? — has no built-in matcher. AnyOfArray is the closest fit but parameterizes its inner matchers on the container's value_type, so EXPECT_THAT("a", AnyOfArray(vector<string>{...})) fails to compile because the subject must be the same (or strictly implicitly-convertible) type.

This adds three public matchers in mbo/testing/matchers.h, all using the outer/Impl pattern already established by CapacityIsMatcher / WhenTransformedByMatcher:

• IsElementOf(container) — value-on-the-left matcher. Comparison goes through a testing_internal::ElementEqual helper that uses raw operator== when the two sides are directly equality-comparable, and falls back to component-wise comparison for std::pair. This lets const char* / std::string_view / std::string subjects all match against a std::vector<std::string> (or std::map<std::string, ...> keys, etc.) without caller-side conversion, and lets std::pair<const char*, const char*> match against std::list<std::pair<std::string, std::string>> even on libstdc++ (C++20 narrowed std::pair::operator== to same-types-only; libc++ accepts heterogeneous via implicit pair conversion, libstdc++ does not).

• IsKeyOf(map) — convenience matcher: "is this value among the keys of map?". Failure messages say "is a key of {1, 2}" / "is not a key of {1, 2}".

• IsValueOf(map) — parallel to IsKeyOf: "is this value among the mapped values of map?". Failure messages say "is a value of {\"a\", \"b\"}" / "is not a value of {\"a\", \"b\"}".

AllKeys / AllValues (the eager std::vector<key_type> / std::vector<mapped_type> projections) are kept in the testing_internal namespace, not exposed publicly. The shapes they enable — IsElementOf(AllKeys(m)) and IsElementOf(AllValues(m)) — are exactly IsKeyOf / IsValueOf, and keeping the projections internal prevents them from being misused as left-side subjects (EXPECT_THAT(AllKeys(m), ...)), which would invert the orientation the matcher exists to provide.

Description leads

IsElementOfMatcher carries configurable positive / negative description leads so failure messages reflect which projection was applied:

• IsElementOf(c) → "is element of {…}" / "is not element of {…}"
• IsKeyOf(m) → "is a key of {…}" / "is not a key of {…}"
• IsValueOf(m) → "is a value of {…}" / "is not a value of {…}"

.trunk/trunk.yaml adjustments

• Drops clang-format@19.1.6: pre-commit's mirrors-clang-format already pins 19.1.6 and works on every platform, while trunk.io's CDN does not host the macOS-arm64 binary for that version. Removing it from trunk eliminates the install failure and the version-split risk; CI's existing pre-commit job continues to enforce formatting.
• Drops clang-tidy@19.1.7 → clang-tidy@16.0.3: trunk.io's CDN also doesn't host clang-tidy-19.1.7-linux-x86_64.tar.gz. 16.0.3 is the plugin's known_good_version and is available on every platform. Pre-commit has no clang-tidy hook, so trunk remains the sole runner; only the version moves.

Test plan

• bazel test //mbo/testing:matchers_test — new tests pass locally (macOS-arm64, libc++)
• Heterogeneous subject types verified: const char[], std::string_view, std::string against std::vector<std::string> and std::map<std::string, ...>
• Pair-of-string-literals subject against std::list<std::pair<std::string, std::string>> — exercises the ElementEqual pair fallback (relevant on libstdc++)
• Edge cases: empty containers, multimap (duplicate keys preserved), unordered_map (hash-order-independent)
• Exact Describe / DescribeNegation / MatchAndExplain strings asserted for IsElementOf, IsKeyOf, and IsValueOf (including empty-map edge cases)
• CI green across pre-commit, trunk, and test-gcc (linux + libstdc++)

🤖 Generated with Claude Code