Add flat_hash_map benchmarks

[lihasgupta]

Adds benchmark suite for absl::flat_hash_map, the recommended default hash map that previously had no public benchmarks.

Benchmarks included:

• Basic operations (insert, lookup hit/miss) for int and string keys
• Load factor analysis at 25%, 50%, 75%, and 87% capacity
• Iteration performance at different load factors
• Reserve vs no-reserve comparison across multiple sizes

Key findings from benchmark results:

• reserve() provides 20-24% performance improvement for known sizes
• Load factor has minimal impact on lookup speed (~2.1-2.2M ops/sec)
• Higher load factors improve iteration by 30% (better cache locality)
• Failed lookups are 50% faster than successful lookups

Testing: All benchmarks compile and run successfully with Bazel.

[lihasgupta]

Hi team,

Polite follow-up on this PR. The benchmarks show some interesting findings on reserve() performance and load factor impacts.

Happy to make any changes or address feedback.

[derekmauro]

Hi and thanks for your interest in Abseil.

We actually do have benchmarks for flat_hash_map, but they are a little hard to find.
https://github.com/abseil/abseil-cpp/blob/master/absl/container/internal/raw_hash_set_benchmark.cc
https://github.com/abseil/abseil-cpp/blob/master/absl/container/internal/raw_hash_set_probe_benchmark.cc

Since we spend so many cycles across our datacenters in this code, it is pretty critical that actual performance changes are validated on production workloads, not microbenchmarks. See https://abseil.io/fast/39.

[lihasgupta]

Hi and thanks for your interest in Abseil.

We actually do have benchmarks for flat_hash_map, but they are a little hard to find. https://github.com/abseil/abseil-cpp/blob/master/absl/container/internal/raw_hash_set_benchmark.cc https://github.com/abseil/abseil-cpp/blob/master/absl/container/internal/raw_hash_set_probe_benchmark.cc

Since we spend so many cycles across our datacenters in this code, it is pretty critical that actual performance changes are validated on production workloads, not microbenchmarks. See https://abseil.io/fast/39.

@derekmauro

Thanks for the clarification! This makes total sense - I should have checked the internal/ directory more carefully.

I appreciate you taking the time to review. I'll close this PR.