Improve char.ToUpperInvariant / ToLowerInvariant perf

[GrabYourPitchforks]

Somewhat inspired by Steve's PR at #35185. Turns out we call char.ToUpperInvariant / char.ToLowerInvariant in a few dozen places within the runtime.

This PR improves these code paths by: (a) not bouncing through the TextInfo.Invariant indirection for ASCII data or when GlobalizationMode.Invariant is enabled; (b) inlining small methods that the JIT wasn't already inlining; and (c) tweaking the bit twiddling to get more efficient codegen.

Method	Toolchain	args	Mean	Error	StdDev	Ratio	Code Size
CharToUpperInvariant	master	Hello World!	51.98 ns	0.615 ns	0.545 ns	1.00	147 B
CharToUpperInvariant	textinfo	Hello World!	22.71 ns	0.436 ns	0.408 ns	0.44	105 B
CharToLowerInvariant	master	Hello World!	49.15 ns	0.295 ns	0.247 ns	1.00	144 B
CharToLowerInvariant	textinfo	Hello World!	22.68 ns	0.242 ns	0.215 ns	0.46	105 B

[Benchmark]
[Arguments("Hello World!")]
public int CharToUpperInvariant(string args)
{
    int accum = 0;
    for (int i = 0; i < args.Length; i++)
    {
        accum += char.ToUpperInvariant(args[i]);
    }
    return accum;
}

[Benchmark]
[Arguments("Hello World!")]
public int CharToLowerInvariant(string args)
{
    int accum = 0;
    for (int i = 0; i < args.Length; i++)
    {
        accum += char.ToLowerInvariant(args[i]);
    }
    return accum;
}

[GrabYourPitchforks]

I also tried testing a branchless implementation but the results were about the same for the different inputs I tried. So no need to complicate the logic.

[gfoidl]

Should the above benchmarks be added to dotnet/performance?
For string there are benchmarks for ToUpperInvariant, but not for char.

[GrabYourPitchforks]

Good catch @gfoidl on the benchmarks repo! I'll get that PR up before closing this out.