Fix string case-insensitive comparisons in Globalization invariant mode

[tarekgh]

This issue addressing #106828.

In Globalization invariant mode when comparing string with case insensitive option, should satisfy the rule: if s1 < s2 and s2 < s3, then s1 < s3. The change here is fixing the string comparison to ensure this rule.

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/area-system-globalization
See info in area-owners.md if you want to be subscribed.

[tarekgh]

@ericstj I am willing to port this fix to 9.0 release, the comment #106828 (comment) is helpful why I want to do that. The change is small and safe with low risk.

[ericstj]

I'm surprised we didn't have cases where this returned the wrong result entirely, for instance when comparing a single upper case to lower case. Do we need more test cases?

[tarekgh]

This case scoped to the globalization invariant mode. The test I added here should be good enough to catch the cases we need to fix.

Note, we have the right implementation in the other places, like

runtime/src/libraries/System.Private.CoreLib/src/System/Globalization/Ordinal.cs

Line 52 in 46c9a4f

return currentA - currentB;

and

runtime/src/libraries/System.Private.CoreLib/src/System/Globalization/OrdinalCasing.Icu.cs

Line 231 in 46c9a4f

return aUpper - bUpper;

[ericstj]

I was just imagining that it might give the wrong answers (let alone ordering) for things like:
a B -> 97 - 65 = 32 <-- incorrect
A b -> 65 - 98 = -33
A B -> 65 - 66 = -1

And I was surprised we didn't have a theory test somewhere for such basic comparisons.

[tarekgh]

Yes, I understand what you are saying but does it really matter much testing single character string or a string with multiple characters as the case I added? if you think this is worth adding I can do it.

[tarekgh]

/backport to release/9.0

[github-actions]

Started backporting to release/9.0: https://github.com/dotnet/runtime/actions/runs/10689249576