Update MessageDifferencer repeated set benchmarks for high-fidelity production modeling.

[copybara-service]

Update MessageDifferencer repeated set benchmarks for high-fidelity production modeling.

Replace the flawed legacy BM_RepeatedFieldSet benchmark with a comprehensive suite of microbenchmarks (BM_RepeatedSet_*) that model real production serving conditions:

• Pre-populate proto payloads and resolve field descriptors outside the timing loop.
• Remove string reporters to evaluate core differencer serving mode (reporter_ == nullptr).
• Cover 3 comparison scenarios: shuffled permutations (0% prefix match), identical ordered sets (guardrail), and single disjoint/mismatched sets.
• Span 4 field types: primitive int32 scalars, SSO strings (<=15 bytes), heap-allocated long strings (>15 bytes), and nested submessages.
• Use realistic production cardinalities: N = 5, 20, 100, 500.