Replace khash + tempita with khashl

[WillAyd]

supersedes #56432

this is just a research project at the moment, but the main thing this solves is using templated functions to achieve better code organization and runtime performance, compared to what's in main.

Haven't spent a ton of time optimization but first benchmark run shows factorization of a high cardinality column can be over 2x as fast, though factorizationn of a low cardinality column is 2x as slow.

| Change   | Before [58e63ec1] <khashl-usage~7^2>   | After [23fc5842] <khashl-usage>   |   Ratio | Benchmark (Parameter)                                                                        |
|----------|----------------------------------------|-----------------------------------|---------|----------------------------------------------------------------------------------------------|
| +        | 1.39±0.04ms                            | 3.05±0.3ms                        |    2.19 | hash_functions.Unique.time_unique_with_duplicates('Int64')                                   |
| +        | 21.4±0.8μs                             | 40.1±20μs                         |    1.88 | hash_functions.NumericSeriesIndexingShuffled.time_loc_slice(<class 'numpy.float64'>, 10000)  |
| +        | 2.55±0.09ms                            | 3.96±0.2ms                        |    1.55 | hash_functions.Unique.time_unique('Int64')                                                   |
| +        | 53.5±3μs                               | 81.6±20μs                         |    1.52 | hash_functions.NumericSeriesIndexingShuffled.time_loc_slice(<class 'numpy.float64'>, 100000) |
| +        | 3.88±0.3ms                             | 5.85±0.8ms                        |    1.51 | hash_functions.Unique.time_unique_with_duplicates('Float64')                                 |
| +        | 15.3±0.2μs                             | 17.2±0.6μs                        |    1.12 | hash_functions.NumericSeriesIndexing.time_loc_slice(<class 'numpy.float64'>, 10000)          |
| -        | 8.46±0.3ms                             | 7.28±0.4ms                        |    0.86 | hash_functions.UniqueAndFactorizeArange.time_unique(5)                                       |
| -        | 9.01±0.2ms                             | 7.48±0.2ms                        |    0.83 | hash_functions.UniqueAndFactorizeArange.time_unique(15)                                      |
| -        | 8.68±0.4ms                             | 7.14±0.3ms                        |    0.82 | hash_functions.UniqueAndFactorizeArange.time_unique(4)                                       |
| -        | 8.84±0.4ms                             | 7.23±0.1ms                        |    0.82 | hash_functions.UniqueAndFactorizeArange.time_unique(9)                                       |
| -        | 8.86±0.4ms                             | 7.14±0.4ms                        |    0.81 | hash_functions.UniqueAndFactorizeArange.time_unique(11)                                      |
| -        | 8.76±0.3ms                             | 6.64±0.1ms                        |    0.76 | hash_functions.UniqueAndFactorizeArange.time_unique(8)                                       |
| -        | 14.4±1ms                               | 7.10±0.8ms                        |    0.49 | hash_functions.UniqueAndFactorizeArange.time_factorize(15)                                   |
| -        | 14.7±0.9ms                             | 6.66±0.3ms                        |    0.45 | hash_functions.UniqueAndFactorizeArange.time_factorize(10)                                   |
| -        | 14.7±1ms                               | 6.58±0.1ms                        |    0.45 | hash_functions.UniqueAndFactorizeArange.time_factorize(11)                                   |
| -        | 15.1±2ms                               | 6.76±0.2ms                        |    0.45 | hash_functions.UniqueAndFactorizeArange.time_factorize(12)                                   |
| -        | 15.4±1ms                               | 6.95±0.3ms                        |    0.45 | hash_functions.UniqueAndFactorizeArange.time_factorize(7)                                    |
| -        | 15.2±1ms                               | 6.92±0.4ms                        |    0.45 | hash_functions.UniqueAndFactorizeArange.time_factorize(8)                                    |
| -        | 14.9±1ms                               | 6.60±0.2ms                        |    0.44 | hash_functions.UniqueAndFactorizeArange.time_factorize(4)                                    |
| -        | 15.4±1ms                               | 6.69±0.1ms                        |    0.44 | hash_functions.UniqueAndFactorizeArange.time_factorize(5)                                    |
| -        | 15.3±1ms                               | 6.77±0.2ms                        |    0.44 | hash_functions.UniqueAndFactorizeArange.time_factorize(6)                                    |
| -        | 15.7±0.8ms                             | 6.27±0.1ms                        |    0.4  | hash_functions.UniqueAndFactorizeArange.time_factorize(9)                                    |

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.
PERFORMANCE DECREASED.

[mroeschke]

Oops accidentally approved

accidental

[github-actions]

This pull request is stale because it has been open for thirty days with no activity. Please update and respond to this comment if you're still interested in working on this.

[WillAyd]

Will reopen if I get time to look at this again. Generally the idea is promising just takes some time to get the benchmarks right