Use memcpy / realloc more effectively in hashtable

[WillAyd]

No description provided.

[WillAyd]

The multiplying by 4 seems somewhat arbitrary and counter-productive for large arrays, but keeping status quo for now

[WillAyd]

There are still a few other places where we are doing loop assignment in the object class, but I have not touched them because I don't know what Cython does for reference counting during assignment that a simple mempcy would not know about. Something to research futher

[WillAyd]

With the extend I choose not to multiple by a factor of 4. This would favor performance when doing really large appends versus many small appends in a loop

Not sure how to best manage this. Long term I would love to get rid of this self-managed code and use the C++ standard library

[WillAyd]

Hashing benchmarks:

| Change   | Before [710720e6] <main>   | After [cfa5b85f] <memcpy-extend>   |   Ratio | Benchmark (Parameter)                                                                        |
|----------|----------------------------|------------------------------------|---------|----------------------------------------------------------------------------------------------|
| -        | 1.12±0.2ms                 | 901±40μs                           |    0.8  | hash_functions.NumericSeriesIndexingShuffled.time_loc_slice(<class 'numpy.uint64'>, 1000000) |
| -        | 1.26±0.1ms                 | 893±30μs                           |    0.71 | hash_functions.NumericSeriesIndexingShuffled.time_loc_slice(<class 'numpy.int64'>, 1000000)  |

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.
PERFORMANCE INCREASED.

algorithms:

| Change   | Before [710720e6] <main>   | After [cfa5b85f] <memcpy-extend>   |   Ratio | Benchmark (Parameter)                                                       |
|----------|----------------------------|------------------------------------|---------|-----------------------------------------------------------------------------|
| +        | 2.11±0.04μs                | 2.34±0.1μs                         |    1.11 | algorithms.Duplicated.time_duplicated(True, 'last', 'duration[s][pyarrow]') |
| -        | 5.81±0.2ms                 | 5.28±0.08ms                        |    0.91 | algorithms.Hashing.time_series_string                                       |
| -        | 2.27±0.2ms                 | 1.89±0.04ms                        |    0.83 | algorithms.Quantile.time_quantile(1, 'linear', 'float64')                   |
| -        | 1.09±0.04ms                | 850±20μs                           |    0.78 | algorithms.Quantile.time_quantile(1, 'nearest', 'uint64')                   |
| -        | 1.89±0.2ms                 | 1.31±0.06ms                        |    0.7  | algorithms.Quantile.time_quantile(0, 'midpoint', 'uint64')                  |

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.
PERFORMANCE DECREASED.

groupby:

| Change   | Before [710720e6] <main>   | After [cfa5b85f] <memcpy-extend>   |   Ratio | Benchmark (Parameter)                                                                   |
|----------|----------------------------|------------------------------------|---------|-----------------------------------------------------------------------------------------|
| +        | 28.5±0.1ms                 | 31.7±1ms                           |    1.11 | groupby.GroupByCythonAggEaDtypes.time_frame_agg('Float64', 'var')                       |
| -        | 29.0±2ms                   | 26.4±0.4ms                         |    0.91 | groupby.GroupByCythonAggEaDtypes.time_frame_agg('Int32', 'last')                        |
| -        | 85.1±10μs                  | 77.1±3μs                           |    0.91 | groupby.GroupByMethods.time_dtype_as_group('int16', 'nunique', 'direct', 1, 'cython')   |
| -        | 24.9±0.7μs                 | 22.6±0.5μs                         |    0.91 | groupby.GroupByMethods.time_dtype_as_group('uint', 'bfill', 'direct', 1, 'cython')      |
| -        | 35.8±3μs                   | 32.1±0.2μs                         |    0.9  | groupby.GroupByMethods.time_dtype_as_group('int', 'std', 'direct', 1, 'cython')         |
| -        | 37.3±1μs                   | 33.7±2μs                           |    0.9  | groupby.GroupByMethods.time_dtype_as_group('uint', 'std', 'direct', 1, 'cython')        |
| -        | 1.33±0.05ms                | 1.19±0.02ms                        |    0.9  | groupby.MultipleCategories.time_groupby_transform                                       |
| -        | 16.9±0.2ms                 | 15.1±0.2ms                         |    0.89 | groupby.GroupByMethods.time_dtype_as_group('uint', 'describe', 'direct', 1, 'cython')   |
| -        | 73.9±1μs                   | 66.0±0.6μs                         |    0.89 | groupby.GroupByMethods.time_dtype_as_group('uint', 'diff', 'direct', 1, 'cython')       |
| -        | 398±5μs                    | 355±2μs                            |    0.89 | groupby.GroupByMethods.time_dtype_as_group('uint', 'pct_change', 'direct', 1, 'cython') |
| -        | 4.54±0.2ms                 | 4.01±0.1ms                         |    0.88 | groupby.Nth.time_groupby_nth_all('datetime')                                            |
| -        | 36.0±2μs                   | 30.9±1μs                           |    0.86 | groupby.GroupByMethods.time_dtype_as_group('int16', 'first', 'direct', 1, 'cython')     |
| -        | 9.09±0.5ms                 | 7.81±0.2ms                         |    0.86 | groupby.Nth.time_series_nth_all('datetime')                                             |
| -        | 618±100μs                  | 394±30μs                           |    0.64 | groupby.GroupByMethods.time_dtype_as_group('float', 'sum', 'direct', 1, 'numba')        |

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.
PERFORMANCE DECREASED.

Not why there are some regressions with algorithms - those may just be flaky tests. I do get different results every time I run the asvs there

[WillAyd]

Alright sorry for letting this go stale for a while but all green and posted updated benchmarks. @mroeschke

[mroeschke]

The performance improvements look great. Feel free to merge once you're ready to take the PR out of draft mode