Optimize BigInteger.ToString for large decimal string

[kzrnm]

This PR is a counterpart to #55121. divide-and-conquer algorithm

Number.FormatBigInteger() can run in $D(n)log(N)$ time using the Divide and Conquer algorithm, where $D(n)$ represents the computational complexity of BigInteger division.

The computational complexity of division have been improved by #96895.

---

Benchmark

Code

using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Configs;
using System.Numerics;

[DisassemblyDiagnoser]
[GroupBenchmarksBy(BenchmarkLogicalGroupRule.ByMethod)]
public class ToStringTest
{
    [Params(100, 1000, 10000, 100000)]
    public int N;

    BigInteger b;

    [GlobalSetup]
    public void Setup()
    {
        b = BigInteger.Parse(new string('9', N));
    }

    [Benchmark] public string DecimalString() => b.ToString();
}

BenchmarkDotNet v0.13.12, Windows 11 (10.0.26100.3037)
13th Gen Intel Core i5-13500, 1 CPU, 20 logical and 14 physical cores
.NET SDK 10.0.100-alpha.1.25077.2
  [Host]   : .NET 10.0.0 (10.0.25.7313), X64 RyuJIT AVX2
  ShortRun : .NET 10.0.0 (42.42.42.42424), X64 RyuJIT AVX2

Job=ShortRun  IterationCount=3  LaunchCount=1  
WarmupCount=3  

Method	Toolchain	N	Mean	Error	StdDev	Ratio	RatioSD	Code Size	Gen0	Gen1	Gen2	Allocated	Alloc Ratio
DecimalString	\main\corerun.exe	100	162.6 ns	72.78 ns	3.99 ns	1.00	0.00	3,976 B	0.0176	-	-	224 B	1.00
DecimalString	\pr\corerun.exe	100	162.2 ns	80.65 ns	4.42 ns	1.00	0.05	3,836 B	0.0176	-	-	224 B	1.00
DecimalString	\main\corerun.exe	1000	10,087.1 ns	1,262.45 ns	69.20 ns	1.00	0.00	3,801 B	0.1526	-	-	2024 B	1.00
DecimalString	\pr\corerun.exe	1000	6,181.8 ns	877.68 ns	48.11 ns	0.61	0.01	3,757 B	0.1602	-	-	2024 B	1.00
DecimalString	\main\corerun.exe	10000	1,042,496.9 ns	207,485.10 ns	11,372.96 ns	1.00	0.00	3,801 B	-	-	-	20026 B	1.00
DecimalString	\pr\corerun.exe	10000	336,682.2 ns	132,151.44 ns	7,243.67 ns	0.32	0.01	3,758 B	1.4648	-	-	20025 B	1.00
DecimalString	\main\corerun.exe	100000	100,113,761.1 ns	18,888,052.58 ns	1,035,317.90 ns	1.00	0.00	3,800 B	-	-	-	200155 B	1.00
DecimalString	\pr\corerun.exe	100000	11,684,286.5 ns	508,446.11 ns	27,869.65 ns	0.12	0.00	3,757 B	46.8750	46.8750	46.8750	200088 B	1.00

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/area-system-numerics
See info in area-owners.md if you want to be subscribed.

[huoyaoyuan]

Does it worth to do a FP calculation, instead of conservatively allocate for upper limit?

[kzrnm]

There's not much difference, but the required storage size will be smaller. For example, when the bit length is 58 or 59, stackalloc is used.

_bit.Length	old	new
57	64	62
58	65	63
59	66	64

This change also aims to achieve mathematical correctness in the implementation.