LGTM.

Do you think it will be valuable to extend this to std::array, std::deque, std::list and std::forward_list at some point (to the extent that they have similar emplace methods) - or is vector::emplace the vast majority of usage anyway?

@dbartol in our meeting last week you asked us to try and avoid merging changes to taint tracking while you finished getting a big PR merged. Is that done, and if not is this change likely to conflict?

I haven't looked closer at the other classes that we don't have support for, but at lest for std::set, std::vector and std::map emplace has slightly different semantics (arguments, return value), so our implementation does not generalize.
The question about usage could be best answered by a query run across LGTM I guess?

Let's wait with merging this until #4432 lands, as we do touch the same test files.

It looks like there are a few uses of the others, particularly list and deque, but not as many as for vector. There are also a lot of uses on a class _Hashtable, which appears to be a commonly occurring implementation class(?). So modelling these is not as important as the vector and map work.