> Every country has a mammoth collection of valid results for your query.
Having seen the corpus of content available from each language, this is categorically false. Consider Wikipedia, which is a fairly ubiquitous information source on the web that provides answers to tons of searches. English documents: 5.4M, Romanian documents: 376k.
Perhaps the solution to the OP's woes is more tools for filtering. This post conflates the ideas between language and country. Search typically returns to you results and search features that are in the query language. Today's solution where filtering is done through query refinement and query operators seems to cover a lot of use cases already.
Further, by having an integrated product, ML models can learn behaviors specific to certain locales where those behaviors differ from region to region, and balance with more universal behaviors that apply to more than one region.