Use ILGenerator instead of Expression

[Nyrest]

Expected & Implemented

• Faster constructor
• Smaller assembly size
• No runtime performance penalty

Issues

• ILGenerator is slower than Expression on .NET Framework.
  • We should use Expression on .NET FX and use ILGenerator on .NET Core

Benchmark result of the static constructors:

BenchmarkDotNet=v0.13.1, OS=Windows 10.0.22000
AMD Ryzen 9 3900X, 1 CPU, 24 logical and 12 physical cores
.NET SDK=7.0.100-preview.1.22110.4
  [Host]     : .NET 6.0.3 (6.0.322.12309), X64 RyuJIT
  Job-RVXJKB : .NET 5.0.15 (5.0.1522.11506), X64 RyuJIT
  Job-CJDGIM : .NET 6.0.3 (6.0.322.12309), X64 RyuJIT
  Job-BTQWWL : .NET Framework 4.8 (4.8.4470.0), X64 RyuJIT

Method	Job	Runtime	Mean	Error	StdDev	Ratio	RatioSD	Gen 0	Gen 1	Allocated
CCtorIL	Job-RVXJKB	.NET 5.0	35.67 μs	0.266 μs	0.207 μs	0.92	0.01	0.1221	0.0610	1 KB
CCtorExpr	Job-RVXJKB	.NET 5.0	38.83 μs	0.364 μs	0.340 μs	1.00	0.00	0.4272	0.1831	4 KB
CCtorIL	Job-CJDGIM	.NET 6.0	31.79 μs	0.501 μs	0.469 μs	0.93	0.02	0.1221	0.0610	1 KB
CCtorExpr	Job-CJDGIM	.NET 6.0	34.03 μs	0.244 μs	0.216 μs	1.00	0.00	0.4272	0.1831	4 KB
CCtorIL	Job-BTQWWL	.NET Framework 4.8	46.18 μs	0.214 μs	0.201 μs	0.99	0.01	0.6104	0.3052	4 KB
CCtorExpr	Job-BTQWWL	.NET Framework 4.8	46.44 μs	0.212 μs	0.199 μs	1.00	0.00	0.6104	0.3052	4 KB

[Nyrest]

For some unknown reason, ILEmit is always slower than Expression.
This is obviously not worth switching to ILGenerator.

This will need more investigation.

Resolved.

[Nyrest]

It seems the DynamicMethod needs to be bound to an object to get the best performance