[Request] Make `unordered_map` as fast as Dinkumware's implementation

Looking at the benchmarks at the [beta docs](https://www.boost.org/doc/libs/1_80_0_beta1/libs/unordered/doc/html/unordered.html#buckets_fast_closed_addressing_implementation), it looks like the two Erasure With Duplicate elements cases against Visual Studio 2019 + Dinkumware were slower. It would be great if this could be addressed for the v1.80 final release.

Thanks!