GH-44237: [C#] Use stack allocated buffer when serializing decimal values

[georgevanburgh]

Span overrides for decimal.GetBits were added in .NET 5, and allow us to avoid a heap allocation for each decimal value serialized.

Running a quick benchmark shows the expected reduction in allocations

using BenchmarkDotNet.Attributes;

namespace Apache.Arrow.Benchmarks;

[MemoryDiagnoser]
public class DecimalUtilityBenchmark
{
    public decimal Value => 1.00000000m;

    private byte[] _buffer = new byte[16];

    [Benchmark(Baseline = true)]
    public void Baseline() => DecimalUtilityMain.GetBytes(Value, 34, 10, 16, _buffer);

    [Benchmark]
    public void Candidate() => DecimalUtility.GetBytes(Value, 34, 10, 16, _buffer);
}

BenchmarkDotNet v0.14.0, Windows 10 (10.0.19045.4894/22H2/2022Update)
13th Gen Intel Core i7-13800H, 1 CPU, 20 logical and 14 physical cores
.NET SDK 8.0.304
  [Host]     : .NET 8.0.8 (8.0.824.36612), X64 RyuJIT AVX2
  DefaultJob : .NET 8.0.8 (8.0.824.36612), X64 RyuJIT AVX2

Method	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Allocated	Alloc Ratio
Baseline	162.7 ns	3.25 ns	4.23 ns	1.00	0.04	0.0088	112 B	1.00
Candidate	152.8 ns	1.38 ns	1.22 ns	0.94	0.02	0.0057	72 B	0.64

Happy to check in some benchmarks for DecimalUtility if they might be useful in future, but thought I'd leave them out for now.

If merged, will close #44237.

• GitHub Issue: [C#] Use new decimal.GetBits overloads to reduce allocations when serializing decimal values #44237

[github-actions]

⚠️ GitHub issue #44237 has been automatically assigned in GitHub to PR creator.

[CurtHagenlocher]

Thanks for the improvement!

[conbench-apache-arrow]

After merging your PR, Conbench analyzed the 4 benchmarking runs that have been run so far on merge-commit dcf0d8c.

There were no benchmark performance regressions. 🎉

The full Conbench report has more details. It also includes information about 10 possible false positives for unstable benchmarks that are known to sometimes produce them.