Experiment with showing both Copilot completions and IntelliSense suggestions when the outputs differ, rather than choosing one or overriding the other.

We could allow this behind a setting and do an experiment, and then segment the results by language or we could already investigate enriching the completions API to add "exhaustive" semantics to the result set. i.e. a result set can be marked exhaustive if the language server is certain that it lists the only legal strings that could come next at a position.