Delete unnecessary internal layer of Span globalization helpers

[jkotas]

This makes the code smaller and faster. For example, the following is about 1.4x faster with this change:

ReadOnlySpan<char> s1 = "Hello world";
ReadOnlySpan<char> s2 = "world";
for (int i = 0; i < 100000000; i++) s1.EndsWith(s2, StringComparison.OrdinalIgnoreCase);

Also, I have ported GetCaseCompareOfComparisonCulture code size optimization from #16434 while I was on it because of it fit well with the rest of the changes.

Edit: The implementation had a bug. The improvement for above example is only 1.4x.

[jkotas]

@dotnet-bot test Windows_NT x64 Checked corefx_baseline
@dotnet-bot test Ubuntu x64 Checked corefx_baseline

[jkotas]

cc @benaadams

[stephentoub]

Excellent. Too bad this isn't in 2.1.

[benaadams]

Nice!

[jkotas]

@dotnet-bot test Windows_NT x64 Checked corefx_baseline

[jkotas]

@dotnet-bot test Windows_NT x64 Checked corefx_baseline

[jkotas]

@dotnet-bot test Ubuntu x64 Checked corefx_baseline