Adds more diagnostics benchmarks

[maryamariyan]

This PR adds benchmarks to compare perf for when there are zero, one or 1000 C# diagnostics in a razor document.
Note: The benchmarks are faking a call to roslyn to get the C# diagnostic.

// * Summary *

BenchmarkDotNet=v0.13.2.2052-nightly, OS=Windows 10 (10.0.18363.2274/1909/November2019Update/19H2)
Intel Xeon CPU E5-1650 v4 3.60GHz, 1 CPU, 12 logical and 6 physical cores
.NET SDK=8.0.100-preview.2.23157.25
  [Host]     : .NET 8.0.0 (8.0.23.12803), X64 RyuJIT AVX2
  Job-UVKTRU : .NET 7.0.4 (7.0.423.11508), X64 RyuJIT AVX2
  Job-NLOVLU : .NET Framework 4.8 (4.8.4510.0), X64 RyuJIT VectorSize=256

PowerPlanMode=00000000-0000-0000-0000-000000000000

|      Method |        Job |            Toolchain |    N | FileType |           Mean |        Error |       StdDev |         Median |            Min |            Max |    Gen0 |    Gen1 | Allocated |
|------------ |----------- |--------------------- |----- |--------- |---------------:|-------------:|-------------:|---------------:|---------------:|---------------:|--------:|--------:|----------:|
| Diagnostics | Job-UVKTRU |             .NET 7.0 |    0 |    Small |       680.5 ns |     13.25 ns |     14.72 ns |       676.8 ns |       664.1 ns |       714.9 ns |  0.0105 |       - |    1008 B |
| Diagnostics | Job-NLOVLU | .NET Framework 4.7.2 |    0 |    Small |     1,450.9 ns |     13.63 ns |     10.64 ns |     1,449.4 ns |     1,435.6 ns |     1,466.3 ns |  0.2041 |       - |    1292 B |
| Diagnostics | Job-UVKTRU |             .NET 7.0 |    0 |    Large |       676.7 ns |      8.74 ns |      7.75 ns |       675.3 ns |       664.0 ns |       695.2 ns |  0.0105 |       - |    1008 B |
| Diagnostics | Job-NLOVLU | .NET Framework 4.7.2 |    0 |    Large |     1,417.6 ns |     19.14 ns |     16.96 ns |     1,418.1 ns |     1,380.3 ns |     1,436.1 ns |  0.2041 |       - |    1292 B |
| Diagnostics | Job-UVKTRU |             .NET 7.0 |    1 |    Small |     1,518.9 ns |     20.69 ns |     17.28 ns |     1,520.6 ns |     1,491.4 ns |     1,545.6 ns |  0.0191 |       - |    1888 B |
| Diagnostics | Job-NLOVLU | .NET Framework 4.7.2 |    1 |    Small |     2,961.7 ns |     16.12 ns |     14.29 ns |     2,959.5 ns |     2,939.5 ns |     2,989.1 ns |  0.3433 |       - |    2166 B |
| Diagnostics | Job-UVKTRU |             .NET 7.0 |    1 |    Large |     3,998.0 ns |     44.20 ns |     41.34 ns |     4,018.0 ns |     3,928.0 ns |     4,047.9 ns |  0.0153 |       - |    1888 B |
| Diagnostics | Job-NLOVLU | .NET Framework 4.7.2 |    1 |    Large |     6,723.4 ns |     69.88 ns |     61.95 ns |     6,740.4 ns |     6,521.6 ns |     6,777.4 ns |  0.3357 |       - |    2166 B |
| Diagnostics | Job-UVKTRU |             .NET 7.0 | 1000 |    Small |   587,534.5 ns |  6,349.26 ns |  5,628.45 ns |   586,930.5 ns |   580,090.1 ns |   596,473.4 ns |  4.8828 |  0.9766 |  522385 B |
| Diagnostics | Job-NLOVLU | .NET Framework 4.7.2 | 1000 |    Small | 1,088,137.8 ns | 10,777.28 ns |  9,553.78 ns | 1,083,954.7 ns | 1,078,160.4 ns | 1,107,036.9 ns | 89.8438 | 25.3906 |  572962 B |
| Diagnostics | Job-UVKTRU |             .NET 7.0 | 1000 |    Large | 3,006,419.8 ns | 30,803.00 ns | 28,813.15 ns | 2,997,595.7 ns | 2,965,953.1 ns | 3,058,587.1 ns |  3.9063 |       - |  522388 B |
| Diagnostics | Job-NLOVLU | .NET Framework 4.7.2 | 1000 |    Large | 4,606,931.6 ns | 30,717.45 ns | 25,650.46 ns | 4,607,103.9 ns | 4,529,810.9 ns | 4,634,828.9 ns | 85.9375 | 23.4375 |  573006 B |

Related to: #8539

TODO next:

• Add a benchmark that also includes the real cost of making an LSP call to roslyn (rather than faking the ClientNotifierServiceBase)
• Add a benchmark that generates an RZ000X diagnostic (razor rather than CS000X diagnostic).

[davidwengier]

I think it would be better if this method was called in the setup method, and the array passed directly to the ClientNotifierService. At the moment the benchmark is measuring the memory allocated by creating this array, which is likely to mask any improvements, and make identifying real problems harder.

[davidwengier]

LGTM. Would be keen to see updated run results, to know if there is actually as much low hanging fruit here as the initial run suggested, or it was just in the diagnostic creation