On the back of #532, this is a PR with further benchmarks for replacing fnv with rapidhash.

Full disclosure, I'm the author of the rapidhash crate and was checking to see if it could give some easy wins.

Summary of findings:

• at small field/value sizes, FnvHasher is very marginally faster.
• at longer field/value sizes, RapidHasher can be significantly faster (my benchmarking suggests the cutoff is likely at around 8-12 chars).
• this assumes a user is using the standard HashMap for labels — rapidhash and fnv hashmaps benefit from using rapidhash in the vec.rs at most lengths, and I haven't got a good explanation for why a default HashMap doesn't see the same improvement.
• I've only been able to benchmark this on an M1 Max chip, I would be curious to see benchmarks on x86, other Arm architectures, and particularly server platforms.

The benchmarks are written as:

• counter_with_mapped_labels the original tests with very short field names around 4 characters (zwei).
• counter_with_mapped_mid_labels a mid-size label of around 12 chars (midsize_zwei).
• counter_with_mapped_long_labels a longer label of around 24 chars (longer_field_name_zwei).

Benchmarking fnv vs. rapidhash actually entails commenting and uncommenting vec.rs lines 9 and 10.

I would love to hear thoughts or see other benchmarks, otherwise if this is of interest I can benchmark on some AWS instances with different architectures.