[clang] [Clang][Sema] Retain the expanding index for unevaluated type constraints (PR #109518)

[Matheus Izvekov via cfe-commits]

mizvekov wrote:

@zyn0217 

I am having second thoughts on both this PR and the follow up fix https://github.com/llvm/llvm-project/commit/adb0d8ddceb143749c519d14b8b31b481071da77

We shouldn't be adding information that is required for correct substitution into `SubstTemplateTypeParmType `, it's sugar after all and doesn't survive canonicalization. That's the reason we have created `SubstTemplateTypeParmPackType`, that one is canonical and is able to hold out information so we can finish substitution correctly later.

I think neither of these patches should have touched `SubstTemplateTypeParmType`, this is probably an issue with not holding out into `SubstTemplateTypeParmPackType`.

I know I reviewed the follow up patch, but I must have missed it was not touching the 'Pack' version of the node instead.

https://github.com/llvm/llvm-project/pull/109518