benchmark: add metrics evaluator benchmark

[timothyw553]

Summary

This adds a focused benchmark for metrics evaluator performance now that the shared benchmark build scaffolding has landed. The benchmark covers evaluator construction separately from data-file evaluation so reviewers can see setup cost and per-file filtering cost independently.

• Covers both InclusiveMetricsEvaluator and StrictMetricsEvaluator.
• Exercises equality, range, starts-with, small IN, medium IN, and null-or-NaN predicates.
• Scales evaluation across 1, 100, and 10,000 synthetic DataFile objects.
• Registers the benchmark in both CMake and Meson benchmark builds.

Part of #690.

Test Plan

• git diff --check origin/main..HEAD
• Configured CMake benchmark build with -DICEBERG_BUILD_BENCHMARKS=ON
• cmake --build build-bench-ci --target metrics_evaluator_benchmark
• build-bench-ci/src/iceberg/benchmark/metrics_evaluator_benchmark --benchmark_filter=BM_InclusiveMetricsEvaluatorMake/equal_id --benchmark_min_time=0.01s
• cmake --build build
• ctest --test-dir build --output-on-failure
• Meson benchmark configure was not run locally because meson is not installed on this machine.

Verification Commands

cmake -S . -B build-bench-ci -G Ninja \
  -DICEBERG_BUILD_BENCHMARKS=ON
cmake --build build-bench-ci --target metrics_evaluator_benchmark
build-bench-ci/src/iceberg/benchmark/metrics_evaluator_benchmark \
  --benchmark_filter=BM_InclusiveMetricsEvaluatorMake/equal_id \
  --benchmark_min_time=0.01s
cmake --build build
ctest --test-dir build --output-on-failure