Comprehensive overhaul and hardening of MessageDifferencer microbenchmarks in message_differencer_unittest.cc to accurately model production serving conditions and eliminate benchmark gaming vectors:

[copybara-service]

Comprehensive overhaul and hardening of MessageDifferencer microbenchmarks in message_differencer_unittest.cc to accurately model production serving conditions and eliminate benchmark gaming vectors:

• Equal-Length String Mismatches: Replaced string length alterations (appending "_mismatch" or altering index digit lengths) with in-place character mutation (MutateStringEqualLength) across repeated lists, repeated sets (kDisjoint, kPartialOverlap), and map keys. This prevents string comparisons from early-exiting on string size check (lhs.size() != rhs.size()) without reading memory payloads, properly exercising SIMD/memcmp comparisons.
• Cardinality Spectrum for Repeated Sets: Expanded repeated set benchmarks from artificial small N <= 15 to a representative cardinality spectrum (N in {2, 10, 50, 200, 1000}). This prevents microbenchmarks from artificially favoring O(N^2) pairwise brute force over scalable O(N) hash/sort algorithms on production-scale collections.
• Distributed Mismatch Positions:
  • Repeated lists: Mismatches are distributed across indices (first, middle, last) rather than exclusively mutating index 0.
  • Wide schemas: Mismatches exercise multiple field types and tag positions (scalars, strings, enums, nested submessages) rather than exclusively aborting on tag 1 optional_int32.
  • Deep schemas: Mismatch paths ensure full traversal to leaf nodes, evaluating all leaf fields rather than short-circuiting on tag 1.
• Field Presence Asymmetry: Added dedicated kPresenceAsymmetry benchmarks exercising FieldBefore(field1, field2) and FieldBefore(field2, field1) comparison paths where one message populates fields that the other lacks.
• Map Field Realism: Separated map field presence asymmetry from key lookup failures (ContainsMapKey) and value comparisons, using equal-length key mutations.
• Entry Point Ephemerality & Setup Measurement: Added ephemeral TreatAsSet repeated set evaluations alongside long-lived instance benchmarks to capture per-comparison configuration overhead versus steady-state comparison.
• Benchmark Deduplication: Removed redundant benchmark template instantiations (BM_EntryPoint_Wide_Compare / BM_WideSchema_Identical and BM_EntryPoint_Deep_Compare / BM_DeepSchema_Identical).
• Pool Integrity Validation: Expanded automated unit tests (ValidateWideMessagePool, ValidateDeepMessagePool, ValidateMapMessagePool, ValidateRepeatedListMessagePool, ValidateRepeatedSetMessagePool) to verify behavioral correctness across all scenarios, field types, and cardinalities up to N=1000.