[clang] [analyzer] Fix crash in PaddingChecker on dependent field types (PR #232986)

[Manuel Sainz de Baranda y Goñi via cfe-commits]

redcode wrote:

> Unless there is urgency - I'd prefer a Sema fix over a workaround. I can also understand that crashing is not really a good outcome while waiting on the proper fix - so this is why I'm asking. Would it work for you all to wait some?

TBH, I don't think the fix does any harm. I'm willing to investigate this further and work on a separate PR where we'd fix Sema, add the unit test you asked for, and add an assert to the guard in `PaddingChecker::shouldSkipDecl` so we don't mask potential future bugs.

But you have more experience than me, so if you think it's not a good idea, I'll close the PR.

https://github.com/llvm/llvm-project/pull/232986