Normalize Regex char class containing a single negated character

[stephentoub]

When RegexFC scans the regex node tree looking for what characters can be the first character, if it encounters a negated character, it handles it by adding everything else to the set. That then ends up defeating optimizations based on IsSingletonInverse, which expects a negated character class containing only the character, as opposed to a non-negated character class containing everything but the character. There is already a RegexCharClass.Canonicalize method; this PR just adds the check and fix-up for this case.

Method	Toolchain	Mean	Error	StdDev	Ratio
Test	\master\corerun.exe	1,087.4 ns	3.47 ns	3.24 ns	1.00
Test	\pr\corerun.exe	425.0 ns	1.43 ns	1.19 ns	0.39

using BenchmarkDotNet.Attributes;
using BenchmarkDotNet.Diagnosers;
using BenchmarkDotNet.Running;
using System.Text.RegularExpressions;

[MemoryDiagnoser]
public class Program
{
    static void Main(string[] args) => BenchmarkSwitcher.FromAssemblies(new[] { typeof(Program).Assembly }).Run(args);

    private readonly Regex _regex = new Regex("[^b]+b", RegexOptions.Compiled);
    private readonly string _input = new string('b', 1000);
    [Benchmark] public bool Test() => _regex.IsMatch(_input);
}

Best reviewed without whitespace: https://github.com/dotnet/runtime/pull/1597/files?w=1

cc: @ViktorHofer, @danmosemsft, @pgovind, @eerhardt

[eerhardt]

Looks good to me. I just had a quick question on one part.