[clang] Correctly compute conversion seq for args to fn with reversed param order (PR #68999)

[Richard Smith via cfe-commits]

https://github.com/zygoloid approved this pull request.

Thanks, looks good.

Our overall approach here seems error-prone, and I wonder if there's a better way to model the reversal of the conversion sequence. For example, perhaps we could change `ConversionSequenceList` into a class that tracks its `OverloadCandidateParamOrder`, and provides both accessors in candidate order (optionally the object parameter, then the remaining declared parameters in order) and in comparison order (potentially reversed). But let's land this as-is before thinking about that :)

https://github.com/llvm/llvm-project/pull/68999