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

[Jared Wyles via cfe-commits]

jaredwy wrote:

> Have you used AI, and read the policies around that? What alternatives have you considered?
> 
> In particular, I'm interested in why should we handle this check at the place where you propose - compared to bailing out earlier - at the top, or not even traversing anything in dependent contexts. So I'm curious of what you think about this.

I think in general we should be doing this, just currently SEMA doesn't do the right thing here as far as I can tell. This seemed a good stop gap, with the fixme and issue. It feel it should bail out at that  if (CXXRD->isDependentType()) check before it. This fix seems ok for now, but we need to revisit why the closure isn't being marked appropriately. 

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