Use hashbrown::RawTable API

[tustvold]

Which issue does this PR close?

Closes #.

Rationale for this change

The use of HashMap without a hasher nor keys, is a little perverse. It turns out hashbrown has a RawTable API that we can use directly instead. As an added bonus this provides a very minor performance improvement

string_dictionary_builder/(dict_size:20, len:1000, key_len: 5)
                        time:   [18.563 µs 18.605 µs 18.647 µs]
                        change: [-4.5058% -3.6392% -2.7326%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 8 outliers among 100 measurements (8.00%)
  3 (3.00%) low mild
  2 (2.00%) high mild
  3 (3.00%) high severe
string_dictionary_builder/(dict_size:100, len:1000, key_len: 5)
                        time:   [21.649 µs 21.693 µs 21.738 µs]
                        change: [-2.2389% -1.8649% -1.5032%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 1 outliers among 100 measurements (1.00%)
  1 (1.00%) high severe
string_dictionary_builder/(dict_size:100, len:1000, key_len: 10)
                        time:   [21.696 µs 21.715 µs 21.734 µs]
                        change: [-2.0519% -1.8956% -1.6960%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 1 outliers among 100 measurements (1.00%)
  1 (1.00%) high severe
string_dictionary_builder/(dict_size:100, len:10000, key_len: 10)
                        time:   [176.06 µs 176.14 µs 176.22 µs]
                        change: [-0.9251% -0.6795% -0.4273%] (p = 0.00 < 0.05)
                        Change within noise threshold.
Found 5 outliers among 100 measurements (5.00%)
  1 (1.00%) low mild
  2 (2.00%) high mild
  2 (2.00%) high severe
string_dictionary_builder/(dict_size:100, len:10000, key_len: 100)
                        time:   [463.94 µs 464.12 µs 464.30 µs]
                        change: [-0.1715% -0.0733% +0.0237%] (p = 0.15 > 0.05)
                        No change in performance detected.
Found 2 outliers among 100 measurements (2.00%)
  1 (1.00%) high mild
  1 (1.00%) high severe

What changes are included in this PR?

Switches to using hashbrown::RawTable in place of HashMap and the raw entry API

Are there any user-facing changes?

No

[tustvold]

Just realised this API is marked experimental

[Dandandan]

I think you need to explicitly enable this feature?

I also used it in the hash join implementation (and I think in hash aggregate as well):
https://github.com/apache/arrow-datafusion/blob/master/datafusion/core/Cargo.toml#L72

[tustvold]

Yeah, I'm just a little bit unsure what the implications of this being an experimental API are... I'll revisit this after #2769